Edition 355 | May 2, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• Why I have been deliberately not shipping a stack of new apps

• The real shift that is coming next, and why it is invisible by design

• What I have actually been building in the background for three and a half years

• The aggregation lane (and the company whose business model points the way)

• The honest paradox of marketing something you are not supposed to notice

• A "Steal This Prompt" to help you start building your own portable context layer this week

Reading Time: 9 minutes Listening Time: 13 minutes

Happy Thursday from Sonoma County.

It is one of those beautiful spring days out here. Sun out. Air warm. The hills are still green from the late rains. I am started writing this from the front lobby of EmpowerED Educational Psychology Services yesterday. Makena is doing her final assessment session, the last piece of the testing process we started a couple of weeks ago. By the end of the month, we will have the full picture.

I will share more about the findings in upcoming editions, including how we are folding the results back into Makena's homeschool curriculum and the custom AI tools we have built around how she actually learns. The plan is to use this real data to keep tightening the fit between her curriculum, her tools, and how her brain wants to work.

This is exactly the kind of feedback loop the rest of this newsletter is about. Real human data. Personalized cognitive partnership. Tools that adapt to the person, not the other way around.

More on that as it unfolds. For today, let me get into what I actually came here to write about.

You have probably noticed something.

For all the categories we have laid out, the frameworks we have published, the prototypes I have shown you behind the scenes, you have not seen LM Lab AI drop a stack of shiny new apps for you to download.

That is not a backlog. That is the strategy.

Here is what I keep coming back to: the last thing the world needs, and the last thing a dyslexic brain needs, is another doom pile of dashboards. Another tab. Another login. Another tool fighting your other tools for a slice of your attention.

For minds like ours, every new app is a tax. Context switching is expensive. Decision fatigue is real. The pile itself becomes the problem we were trying to solve in the first place.

Where This Is Actually Going

I have been telling you for years now, across editions, white papers, and research notes, that the real shift is not more software.

It is agent-to-agent.

Your agent talks to mine. Mine talks to your bank's. Yours coordinates with your kid's school, your doctor, your coach, your accountant. The websites you click through today are going to start feeling weirdly identical, because most of that surface area is collapsing into one conversation: you and your cognitive partner.

That partner will know you. Your patterns. Your kids. Your goals. Your reading speed. Your blind spots. Your strengths. Not because you filled out a 40-field form, but because you have been building that context for years already, in every conversation you have with your AI.

The personalization is not coming. It is already happening. Most people just have not noticed yet.

So Why No Tool From Us?

Because building another app for you to manage would have been exactly the wrong move.

What I have been doing instead, quietly, in the background, is research and testing. The DLM three-layer architecture (Socratic, Strategic, Skeptic). The Cognitive Balance Model. The Triangle Tool. The Single Source of Truth. The whole skills and prompt ecosystem that lets your AI activate the right context, the right way, every session.

That work was never about shipping a SaaS product.

It was about being ready for when the agent era actually arrived. Which is now.

So instead of adding to your pile, the goal is the opposite. To reduce it. To help you build a context layer so personalized that the apps fall away on their own. You do not manage the tools. The tools manage themselves around you.

What We Are Actually Good At

There is a flip side to "we have not shipped a tool" that I want to make explicit, because it is easy to hear that and assume we have not done anything. We have. It is just not the thing people expect from a startup.

What I am genuinely good at, and what the team is good at, is aggregation.

Taking the best of what everyone else is building. Pulling it together. Stress-testing it. Finding the workflow, the prompt structure, the agent architecture that turns five separate tools into one coherent experience that actually fits how a dyslexic brain, or any specific brain, wants to work.

We do not need to rebuild every wheel. We need to know which wheels to use, when, and for whom. That is a different muscle than building a model from scratch, and it is the one I have been training for the last three years across 355 editions.

We have our own creative processes, workflows, and agents in motion. But the real edge is in the assembly, not the invention.

A loose business-model parallel I think about is Perplexity. When they started, the conventional wisdom was that you could not win in AI unless you trained your own frontier model. They went the other way, model-agnostic, ruthlessly good at search, accuracy, and orchestration. The moat was not the model. The moat was knowing how to assemble what was already out there into something more useful.

I am not Perplexity. They have a massive team and a multibillion dollar valuation. But the strategic principle is one I think applies to what I am building. Stay model-agnostic. Stay person-specific. Build the orchestration layer, not the next frontier model.

A truly personal agent should not be locked to one model. It should pick the right tool for the right person, on the right project, at the right moment. The model that works best for how I think is not necessarily the model that works best for someone else doing the exact same job at the same company, because they think differently. Same role. Different brain. Different tool.

That orchestration layer, model-agnostic, person-specific, context-aware, is what we have quietly been building under the hood. Not a single shiny app. The connective tissue between everything else.

The Honest Paradox

Here is the part that is hard to sell.

I am building something you are not supposed to see every day.

If I do this right, you will not open a "dyslexic AI app." You will just live a life where everything is quietly bent toward how your brain actually works. The accommodation is invisible. The friction is gone. The cognitive load drops. The work feels like yours again.

But I still need you to know we exist.

I still need this to be a real business that funds real research. I still need the founding members, the subscribers, the people who read these editions and say yes, I want that future, and I will back the people building toward it.

That is the strange position we are in. Marketing the invisible. Asking you to fund a thing whose endgame is that you barely notice it is there, because everything finally just works.

I think it is worth it. After 355 editions, I am pretty sure you do too.

OK But What Do I Actually Do With This?

You do not need to wait for the agent era to start preparing for it. You are already preparing every time you talk to your AI. The question is whether you are doing it on purpose.

Three things this week.

1. Notice your pile. What apps are you using right now that an agent could replace in the next twelve months? Stop emotionally investing in those. Do not learn the new feature. Do not watch the tutorial. Use them lightly.

2. Feed your context, deliberately. Every conversation is training data for your future cognitive partner. Talk to it like the lifelong partner it is becoming, not like a search engine.

3. Document what you wish it already knew. The patterns. The history. The values. The way you actually think. Get those into a portable context document now, while you are early.

Steal This Prompt

A starter prompt to begin building your own portable context layer. The document you will paste into any future AI agent so it understands you in 30 seconds instead of 30 sessions.

Run it. Save the output. That document is the seed of your cognitive partner.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR For My Fellow

🚫 The reason you have not seen LM Lab AI ship another app is the whole strategy. Another tab is a tax on a dyslexic brain. The doom pile of dashboards is the problem, not the solution.

🤝 The real shift is agent-to-agent. Your agent talks to mine. Mine talks to your bank's, your kid's school's, your doctor's. The apps collapse into one conversation between you and your cognitive partner.

🏗️ What I have been building quietly: the orchestration layer, not another app. The DLM three-layer architecture, Cognitive Balance Model, Triangle Tool, Single Source of Truth, and the prompt and skills ecosystem that activates the right context every session.

🧩 The lane is aggregation, not invention. Taking the best of what everyone else is building and assembling it into something that fits how your specific brain actually wants to work. Model-agnostic. Person-specific. Context-aware.

🪞 The honest paradox: I am building something you are not supposed to see every day. If I do this right, you will not open a "dyslexic AI app." You will just live a life where the friction is gone.

📨 Three things to do this week: notice the apps you are emotionally invested in (do not be), feed your AI context deliberately, and document what you wish it already knew.

🔧 The "Steal This Prompt" in this edition will get you started building a portable context document an agent can read in 30 seconds.

Previously

• Edition 354: "I Chose the Hardest Path on Purpose" (the reminder edition, mission and endurance)

• Edition 353: "What I Am Actually Building With LM Lab AI" (the vision, five lanes, agent stack)

• Edition 346: "The Meta Layer" (builder evaluation, cognitive fit, six-tool stack)

• Edition 345: "We Have Been Asking the Wrong Question About AI" (evaluation framework manifesto)

• Edition 344: "I Woke Up at 4AM With a Random AI Idea" (Cognitive Partner OS prototypes)

• Edition 339: "Your AI Just Forgot Everything. Again." (Karpathy, five-layer self-improving stack)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model, HGI)

• Edition 329: "Building Your Second Brain" (Single Source of Truth)

Edition 354 | May 1, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• Why I picked a problem with no cure and no clean solution

• The difference between a quick fix and a long fight worth fighting

• Why I am writing this for myself as much as for you

• How dyslexia and AI both exist on spectrums that will never resolve

• What it means to be a leader in a field that does not have rules yet

• Three things to remember the next time you doubt the path you are on

• A "Steal This Prompt" you can use to build your own mission reminder document

Reading Time: 8 minutes Listening Time: 11 minutes

Happy Tuesday.

I want to write something today that is more for me than for you.

But I have a feeling some of you need to read it too.

The Problem I Picked

Three and a half years ago, I started writing about dyslexia and AI.

I want to be honest about what I was doing.

I picked a problem with no cure. Dyslexia is not something that gets solved. There is no pill. There is no surgery. There is no app that flips a switch and makes a dyslexic brain process text the way a neurotypical brain does. There never will be. Edition 343 ("Stanford Just Measured Everything") referenced research showing the right tools can physically rewire the dyslexic brain. That is real progress. It is also not a cure. It is a different way of moving forward, like the wheelchair analogy I used in Edition 349 ("A New Paper Just Named the Problem I Have Been Writing About for Three Years").

I picked a tool with no finish line. AI is not finishing. Models keep getting more capable. Workflows keep evolving. New categories of products keep emerging. I am writing this on a Friday in early May 2026, and the AI tools I will be using six months from now do not exist yet.

So I picked a problem that will never be solved and a tool that will never stop changing.

On purpose.

Why That Was the Point

I have been a competitive person my whole life. Sports. Business. Building. I played to win.

When I was younger, I thought competitive meant fast. Quick decisions. Fast reps. Speed wins games.

What I have learned is that competitive really means showing up consistently for something that matters when most other people stop showing up. Speed is a tactic. Endurance is a strategy.

The reason I picked dyslexia and AI is not because I thought it would be easy. It is because I knew it would never be done.

Most people are looking for a finished product. A clean answer. A cure. A perfect tool. A definitive framework. A simple guide.

I picked a space where none of those things exist.

Where the work will keep evolving. Where the questions will keep shifting. Where the answers will keep needing to be rebuilt as the technology, the science, and the lived experience all keep moving.

That is not a bug in my plan. That is the whole reason I picked it.

Because I want to do work that keeps me on my toes. Work that requires me to keep learning. Work where the moment I get comfortable is the moment I have stopped doing it well.

If you have been here since Edition 332 ("A Year Ago, I Was in a Hospital Bed"), you know I am writing this from the other side of a hard year. The Cognitive Balance Model came out of that. The Human Guidance Index came out of that. Those frameworks were built in real time, in public, while I was figuring out who I would be on the other side of a serious health scare.

I did not pick easy. I picked work that mattered enough to keep doing.

Why This Is the Reminder I Needed Today

I am writing this on a day where I needed to remember.

Sometimes it gets heavy. The mission is bigger than what I can do in a day. The progress is slow. The audience is growing but the people I most want to reach still feel one step out of reach. The platform is built but the launch is still ahead. The frameworks are real but the bigger application still has not arrived.

In Edition 342 ("The Weight in My Chest"), I wrote about that feeling. The waiting to be understood. The sense that more people should get it by now.

I told you in Edition 342 that the feeling was not frustration. It was fuel.

That is still true. And some days I need to remind myself of that.

Because picking a hard path on purpose does not make the path less hard. It just means you knew what you were getting into.

I knew dyslexia was going to be a permanent battle. I knew AI was going to be a moving target. I knew that combining them would mean three and a half years of building tools that are out of date the day after I finish them.

I picked it anyway.

And on the days I forget why, I have to come back to this.

The Big-System Pattern

The reason I find this work meaningful is because it sits in a category of problems that I think are the most important problems we face right now.

Education. Medicine and pharmaceuticals. The food industry. The energy industry. Fossil fuels and the climate. Healthcare access. The mental health crisis. The accessibility divide we covered in Edition 351 ("Who Is AI Actually Serving Right Now?").

These are big-system problems. They do not have clean solutions. They have powerful incumbents. They have political headwinds. They have entrenched interests that benefit from the status quo. They have decades of inertia.

And yet, they are the problems that matter most.

Most people opt out of these fights because there is no clean victory available. The work feels like pushing on a system that does not want to move.

But these are exactly the fights that need people who can take the long view. People who are willing to make incremental progress without immediate reward. People who can keep showing up when the wins are slow and the losses are personal.

I am not saying I am that person every day. I am saying I am trying to be.

That is what this newsletter is about, at the most honest level. It is the public record of a dyslexic guy from Sonoma County trying to make a small contribution to one of those big-system battles. With the tools that exist. From the lane I was given.

What Leadership Means in a Field With No Rules

I will be honest about the ambition.

I want to lead.

Not in a hierarchical way. Not in a CEO-of-everything way. In a "I want to be the most useful person to the most dyslexic and neurodivergent thinkers I can reach" way.

I want to build the best tools. I want to write the most honest frameworks. I want to share the work openly so that people can copy it, adapt it, and build their own versions. I want to be the person who takes the heat from the front of the line so that the people behind me can move more easily.

That is what leadership looks like in a field that does not have rules yet. It is not about being the smartest. It is about being the most consistent. It is about staying in the conversation when the conversation gets repetitive. It is about pushing forward when the trend cycle moves on.

In Edition 353 ("What I Am Actually Building With LM Lab AI"), I laid out the bigger vision. The five lanes. The agent stack. The cognitive partnership engine. The path from solo entrepreneur to scaled solo operator.

That vision only works if I keep showing up. Not just for the easy weeks. Especially for the hard weeks.

I want to lead the largest community of dyslexic thinkers and neurodivergent professionals using AI well. Not by being above them. By being one of them, working out loud, and refusing to leave anybody behind.

That ambition is hard to say out loud because it sounds bigger than I usually let myself sound. But it is true. And on a day where I need to remember why I am doing this, saying it out loud is part of how I get back on the path.

OK But What Do I Actually Do With This?

Three things. The next time you doubt the path you are on.

1. Ask Yourself What You Picked, and Why

If you are working on something hard right now, take five minutes to write down why you picked it. Not the elevator pitch. The real reason.

You may have picked it because it was hard. You may have picked it because nobody else was. You may have picked it because the easy paths were not available to you. Whatever the reason, it is worth remembering.

The path you chose did not get harder. You just forgot why you chose it.

2. Find Your Long-Term Pattern, Not Your Short-Term Win

Most people measure progress by what happened this week. Long-term work needs long-term metrics.

Look at what you have built over the last six months. Twelve months. Three years. The progress is rarely visible day to day. It is usually visible in the longer arc.

For me, it is 354 editions of this newsletter. For you, it might be the calluses on your hands, the relationships you have built, the work that exists in the world because you kept showing up.

Find your long-term pattern. Trust it.

3. Pick the Reminder You Need

If you have a hard mission, pick a reminder you can come back to on the worst days.

Mine is something like: I picked this because it would keep changing forever, and that was the point.

Yours might be different. Write it down. Put it somewhere you can see it. Read it on the days you need it.

What This Means for You Right Now

I will close with this.

I am writing this on a day where I needed the reminder. The mission is real. The work is hard. The progress is slow. The path is long.

And I picked it on purpose.

If you are reading this and you are also on a long path, working on something that does not have a clean solution, building tools or running a business or raising kids or surviving an illness or trying to make a small dent in a system that does not want to move, I see you.

You picked something hard. You picked it for a reason. The reason still matters even on the days you forget.

Show up tomorrow.

Then show up again the next day.

That is how big-system work gets done. Not by people who solve it. By people who refuse to walk away.

Steal This Prompt

A reminder document for the days you forget why you picked the path you are on. Drop this into Claude or ChatGPT, and let it help you build a personal "why" document you can come back to whenever the work feels heavy.

Save the output. Read it when you need it. Update it once a quarter. The reminder is the work.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR - For My Fellow Skimmers

🛤️ I picked a problem with no cure and a tool with no finish line. On purpose. Most people are looking for a finished product. I picked a space where none of those exist.

⏳ Competitive does not mean fast. It means showing up consistently for something that matters when most other people stop showing up. Speed is a tactic. Endurance is a strategy.

💪 Big-system problems (education, healthcare, energy, climate, accessibility) do not have clean solutions. They have powerful incumbents and entrenched interests. They are also the problems that matter most. They need people who can take the long view.

🧭 Leadership in a field without rules is not about being the smartest. It is about being the most consistent. It is about staying in the conversation when the conversation gets repetitive. It is about pushing forward when the trend cycle moves on.

🎯 I want to lead the largest community of dyslexic thinkers and neurodivergent professionals using AI well. Not by being above them. By being one of them, working out loud, and refusing to leave anybody behind.

🔁 Three things to remember the next time you doubt your path: ask yourself what you picked and why, find your long-term pattern instead of your short-term win, and pick a reminder you can come back to on the worst days.

🔒 Cognitive Partner Members get the live work as it builds. 50 founding spots at $19/month, locked forever. The mission is the long game.

Previously

• Edition 353: "What I Am Actually Building With LM Lab AI" (the bigger vision, five lanes, agent stack)

• Edition 352: "We Hold AI to a Standard We Have Never Held Ourselves To" (mistakes, double standards)

• Edition 351: "Who Is AI Actually Serving Right Now?" (Gen Z survey, accessibility divide)

• Edition 349: "A New Paper Just Named the Problem I Have Been Writing About for Three Years" (LLM Fallacy, wheelchair analogy)

• Edition 343: "Stanford Just Measured Everything About AI" (AI Index, Stanford dyslexia research)

• Edition 342: "The Weight in My Chest" (autonomy, sticktoitness, the mission)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model, HGI[Get the Free Prompt Guide →]

Edition 353 | April 30, 2026

The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• The vision I have been quietly working toward for three and a half years

• Why dyslexic thinkers see this opportunity before most people do

• The five businesses and interests I want to build agents around

• Why the Dyslexic AI Newsletter is my hero's journey but LM Lab AI is the bigger mission

• How a self-written newsletter actually works (and why "self-written" is not what most people think it means)

• Three things you can do this week to start building your own version

Reading Time: 9 minutes

Listening Time: 13 minutes

Happy Monday.

I want to tell you what I am actually building.

Not the surface version. Not the elevator pitch. The real version.

For three and a half years, I have been writing this newsletter and building tools and frameworks under two related but distinct brands: Dyslexic AI and LM Lab AI. They are connected, but they are not the same thing.

Dyslexic AI is my hero's journey. It is the project where I share what it actually looks like to be a dyslexic thinker working with AI in real time. The frameworks. The tools. The insights. The vulnerability. The community. This newsletter you are reading. The Cognitive Partner Membership. The Single Source of Truth. The Cognitive Balance Model. All of it.

LM Lab AI is the bigger mission. It is the work of helping everyone else build their own version of what I am doing for myself. Helping other people, other businesses, other industries discover what cognitive partnership looks like when you build it around how their brain actually works, what their goals actually are, and what their day-to-day life actually looks like.

Today I want to tell you what that bigger mission really looks like. Because I think we are closer to it than most people realize.

The Solo Entrepreneur With the Output of a Company

Here is the vision in one sentence.

A solo entrepreneur, working with a stack of AI agents and workflows tailored to their specific cognitive style, can produce the output and impact of a small company.

That is not a hypothetical. That is a thing that is now possible. Right now. With the tools that exist this week.

The catch is that almost nobody is actually building it that way. Most solo entrepreneurs are still using AI the way most people use AI: as a chatbot. Open the window. Ask a question. Get an answer. Close the window.

That is not the model.

The model is building agents that work in your voice, on your terms, around the things you actually care about. Agents that already know your background. Agents that produce content in your style. Agents that handle the parts of your business that drain you while leaving you free to focus on the parts that make you feel alive.

If you have been here since Edition 344 ("I Woke Up at 4AM With a Random AI Idea"), you know about the Cognitive Partner OS I started prototyping. Edition 345 ("We Have Been Asking the Wrong Question About AI") laid out the evaluation framework. Edition 340 ("I Have Four of the Five Layers") introduced the self-improving loop.

All of those pieces are part of the same vision. They are the pieces that, assembled together, become the operating system for a solo entrepreneur with the output of a company.

What "Self-Written Newsletter" Actually Means

Let me ground this in something concrete.

I want this newsletter to be self-written. And I want to be very clear about what that does and does not mean.

It does not mean an AI cranks out content with no human in the loop.

It does mean an agent drafts in my voice, on subjects I have flagged as priorities, using a Single Source of Truth that knows me well, with me as the editor reviewing, refining, and approving every edition before it ships.

In that future, I am not the writer. I am the editor.

I look at what the agent has drafted on the topics I flagged this week. I read it the way an editor reads a piece submitted by a staff writer. I push back on the parts that do not sound like me. I tighten what needs tightening. I sign off when it is ready.

The voice stays mine because the agent was built on years of my actual voice. The topics stay mine because I picked them. The integrity stays intact because I am still the one approving every word that goes out.

This is not laziness. This is leverage.

Edition 332 ("A Year Ago, I Was in a Hospital Bed") introduced the Cognitive Balance Model. Three phases: Human Initiation, AI Expansion, Human Integration. The self-written newsletter is that model at scale. I initiate the topic. The agent expands it. I integrate by editing and approving.

The output looks like a writer. The reality is an editor. And in a world where editorial judgment is more valuable than writing speed, that is the right place to be.

Why Dyslexic Thinkers See This First

Here is the part that matters.

Dyslexic and neurodivergent thinkers are uniquely positioned to see this opportunity before most people do. And that is not flattery. It is a real observation about cognitive style.

We are pattern thinkers. We see across domains. We connect ideas that other people keep separate. Edition 343 ("Stanford Just Measured Everything") referenced the "jagged frontier" finding. AI is brilliant in some areas and terrible in others. Dyslexic minds have lived on a jagged frontier our whole lives.

We are also iterative thinkers. We do not get things right the first time. We never have. So we developed the ability to look at output, figure out what is off, and try again with better information. That is exactly what running an agent stack requires. Test, refine, evaluate, adjust. Most neurotypical thinkers find that exhausting. We find it natural.

We are also experience-driven. I did not design my cognitive partner work in a research lab. I built it because I needed it. Because every existing tool was hostile to how my brain works. Necessity drove invention.

That same pattern shows up across the dyslexic and neurodivergent community. We are inventors not because we are smarter, but because the world we lived in did not work for us out of the box. So we built our own. And now, finally, the tools exist to scale what we have always done.

In Edition 338 ("Palantir's CEO Just Made It Official"), we covered the case for the neurodivergent advantage in AI work, with Gartner predicting 20% of Fortune 500 companies will actively seek neurodivergent talent. Edition 334 ("The Data Is In") showed the labor market data on cognitive flexibility as job security.

The data is pointing the same direction the experience does. Dyslexic thinkers are early movers on this transition. The question is not whether we are positioned for it. The question is what we build now while we have the lead.

The Five Lanes I Care About

Here is where it gets personal again. Because LM Lab AI is not theoretical for me. It is built on the businesses, industries, and passions I have lived inside for years.

Lane 1: Neurodivergent Thinking. Dyslexia. ADHD. Autism. Lateral thinking. Outside-the-box cognition. This is the lane where Dyslexic AI lives, where this newsletter lives, where the Cognitive Partner Membership lives. The frameworks I have built, the Cognitive Balance Model, the Human Guidance Index, the DLM Three-Layer Architecture, the Single Source of Truth, all start here. This is the personal one.

Lane 2: Recreation Administration and Business. I owned a gym with my ex-wife for close to fifteen years. I helped create a stand-up paddleboard business. I was involved in the early days of stand-up paddleboard instructor associations and professional standards for the industry. Before that, we had a mobile rock climbing wall business. Recreation is in my bones. There is an enormous opportunity to build cognitive partner agents for gym owners, recreation professionals, instructors, outfitters, and small business operators in this space. People who never saw themselves as "tech" but desperately need the leverage AI can provide.

Lane 3: Sonoma County and Northern California Recreation. I live here. I love it here. The wine country, the redwoods, the coast, the rivers, the trails. There is a regional layer to LM Lab AI that focuses on local businesses, local creators, local entrepreneurs, and local communities. People who want AI tools that understand where they live, who their customers are, and what their neighborhoods need. That is a different kind of AI than the global SaaS version most companies are selling.

Lane 4: Tahoe and Outdoor Recreation. Tahoe has my heart for different reasons. Mountain culture. Snow sports. Lake life. Outdoor business. The outdoor recreation industry needs serious help thinking about AI in a way that respects how the work actually happens. A snowboard instructor is not a knowledge worker. A mountain guide is not a SaaS founder. The tools need to fit the work.

Lane 5: Homeschool and Alternative Education. This one is family. My daughter Makena is homeschooled. Edition 325 ("My 14-Year-Old Daughter Just Proved Me Wrong") kicked off the family thread. Edition 337 ("I Gave My Old Mac Mini a Brain") previewed the Homeschool Parent's Guide to AI. The family evaluation tool from Edition 341 is already in motion. There is a whole community of homeschool families, alternative educators, and parents of neurodivergent kids who need this work and cannot get it from the big AI companies.

These five lanes are not separate businesses. They are five expressions of the same underlying engine.

The engine is cognitive partner AI built around the human who uses it. The applications change. The principles stay the same.

Why This Is Different From What Most People Are Building

I want to draw a clean distinction.

Most AI companies are building horizontal products. ChatGPT for everyone. Claude for everyone. Universal tools designed to work for the most general use case possible.

That is fine work. We need those tools. They are the foundation of everything else.

But what is missing, and what dyslexic thinkers are uniquely positioned to build, are vertical, vocabulary-aware, cognitive-style-aware applications that take the foundation models and shape them around specific people, specific industries, specific workflows, and specific brains.

The future I am building toward is not a future where everyone uses the same AI. It is a future where everyone has the right AI for them.

A gym owner in Petaluma should not be using the same AI workflow as a venture-backed SaaS founder in San Francisco. A homeschool parent in Sebastopol should not be using the same AI tools as an enterprise compliance officer. A neurodivergent freelancer should not be using the same prompt patterns as a Fortune 500 marketing team.

They all deserve cognitive partners that fit them. Not generic tools they have to fight with.

That is the LM Lab AI mission. That is what I have been building toward, edition by edition, framework by framework, prototype by prototype.

Where the Newsletter Fits

The Dyslexic AI Newsletter is not the product. It is the proof of concept.

Three and a half years of evidence. 353 editions. Frameworks. Tools. Voice. Community. A demonstration that one dyslexic thinker, working with AI thoughtfully and consistently, can build something that grows in value over time.

When the self-written version of this newsletter ships, it will be the proof that the vision works. Not in theory. In practice. With me as the editor, the agents as the workforce, and the output staying authentically mine.

If I can do that, anybody can do that.

A gym owner could have an agent that drafts their member newsletter, their class descriptions, their marketing emails, and their social content. Not soulless. Not generic. Drafted in their voice, around their gym, for their community, with them as the editor.

A homeschool parent could have an agent that supports their kid's curriculum, drafts week plans, suggests projects, and adapts to how that specific kid actually learns. With the parent as the editor of the experience.

A paddleboard instructor association could have agents that handle certifications, draft regional newsletters, support members, and produce educational content that respects what the instructors actually do.

These are not science fiction. These are the kinds of projects I want LM Lab AI to make possible.

And the only reason I can see this clearly is because I have been building my own version for three and a half years, in public, on this newsletter, with you watching.

OK But What Do I Actually Do With This?

Three things. This week.

1. Identify Your Five Lanes

What are the five areas of your life or work where you have real depth? Not surface knowledge. Real, lived expertise. Hobbies count. Past careers count. Family roles count. Communities count.

For me, it was neurodivergent thinking, recreation business, regional Northern California, Tahoe, and homeschool. Yours will be different. The point is to know where you have already done the cognitive work that an AI agent could be built around.

2. Pick One Lane and Imagine the Agent

Pick one of your lanes. Now imagine a cognitive partner agent designed specifically for that lane. What would it know about you? What would it produce? What would it free you up to do?

You are not building it yet. You are imagining it clearly enough that you could describe it to someone else. That description is the seed of every agent you might eventually build.

3. Tell Me What You See

If this resonates, hit reply. Tell me what your five lanes are. Tell me what your one imagined agent would do. Tell me where you would start if you had the tools.

I am building the tools. I would rather build them with the people who would use them than for the people who might.

That is what LM Lab AI is for. That is what the Cognitive Partner Membership is for. That is what every framework I have written about for 353 editions is for.

You are not just an audience. You are the reason this work exists.

What This Means for You Right Now

I have been writing this newsletter for three and a half years.

Some weeks I wonder if it lands. Some weeks I worry the work is invisible. Some weeks I get the feeling I wrote about in Edition 342 ("The Weight in My Chest"). The feeling of waiting to be understood.

But I keep showing up. Because the vision is real. Because the work matters. Because the people who think differently, who build differently, who see opportunities that the mainstream misses, deserve tools that fit them.

LM Lab AI is the bigger version of that mission. The Dyslexic AI Newsletter is where I prove it can be done. The Cognitive Partner OS is the engine. The evaluation frameworks are the safety rails. The Single Source of Truth is the foundation. The agents are the leverage.

It all connects. It has always connected. And I am closer than I have ever been to making it real.

If you want to be part of it, you already are.

The question is just how big you want your own version to become.

Matt "Coach" Ivey

Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

🎯 Dyslexic AI is my hero's journey. LM Lab AI is the bigger mission. Helping everyone else build their own version of what I have been building for myself for three and a half years.

🤖 The vision in one sentence: a solo entrepreneur, working with a stack of AI agents and workflows tailored to their specific cognitive style, can produce the output and impact of a small company. That is not science fiction. That is now.

✍️ "Self-written newsletter" does not mean no human in the loop. It means agent drafts, human edits, voice stays authentic, output scales. Same Cognitive Balance Model from Edition 332. Just at a new level.

🧠 Dyslexic thinkers are early movers on this transition. We are pattern thinkers. Iterative thinkers. Necessity-driven inventors. The cognitive style that has been undervalued is the one this moment rewards.

🛶 Five lanes I am building around: neurodivergent thinking, recreation business (gym, paddleboarding, rock climbing), Sonoma County and Northern California, Tahoe and outdoor recreation, and homeschool and alternative education. Different applications. Same engine.

🌐 Most AI companies are building horizontal products for everyone. The opportunity is vertical, cognitive-style-aware, vocabulary-aware applications built around specific people, specific industries, and specific brains.

🛠️ Three things to do this week: identify your five lanes (areas of real depth), pick one and imagine the agent that would serve it, then tell me what you see.

🔒 Cognitive Partner Members get the engine, the frameworks, and the live build updates. 50 founding spots at $19/month, locked forever.

Previously

• Edition 352: "We Hold AI to a Standard We Have Never Held Ourselves To" (the mistakes double standard)

• Edition 351: "Who Is AI Actually Serving Right Now?" (Gen Z survey, accessibility divide)

• Edition 350: "MIT Just Taught AI to Say 'I'm Not Sure'" (calibration, ternary thinking)

• Edition 345: "We Have Been Asking the Wrong Question About AI" (evaluation framework manifesto)

• Edition 344: "I Woke Up at 4AM With a Random AI Idea" (Cognitive Partner OS prototypes)

• Edition 342: "The Weight in My Chest" (autonomy, the mission, sticktoitness)

• Edition 340: "I Have Four of the Five Layers" (self-improving loop)

• Edition 338: "Palantir's CEO Just Made It Official" (neurodivergent advantage)

• Edition 337: "I Gave My Old Mac Mini a Brain" (Homeschool Parent's Guide preview)

• Edition 334: "The Data Is In" (cognitive flexibility, labor market)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model, HGI)

• Edition 325: "My 14-Year-Old Daughter Just Proved Me Wrong" (Makena, family AI)

Next

Edition 354: A milestone reflection. We have crossed 350 editions. Three and a half years. I want to write about what has changed, what has not, and what the next 350 should be focused on. Plus a real ask for the community I have built with you.

Edition 352 | April 29, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• Why I worry about sending out a newsletter with mistakes (and why that worry is unfair)

• The customer service argument that flips the AI mistakes debate on its head

• Why "AI hallucinates" is a real concern wrapped in a double standard

• How dyslexic and neurodivergent thinkers see this differently

• Why mistakes have always been part of human creative work, sometimes on purpose

• Three things to do this week when you find an AI error in your own work

Reading Time: 8 minutes Listening Time: 11 minutes

Happy Wednesday.

I have a confession.

I send out this newsletter knowing it might have mistakes in it.

Not glaring ones. Not factual errors I should have caught. But the small ones. A weird sentence that got stitched together when I combined two voice transcripts. A phrasing that sounds slightly off because my dictation jumped between thoughts. A word my AI cleaned up that lost a little bit of my actual meaning.

I am dyslexic. I hate proofreading. I rely on AI to help smooth my dictated transcripts into something readable. I do my best to catch errors. Sometimes I miss them.

And every time I notice one after the fact, the same feeling shows up.

It makes me feel unauthentic.

Like the mistake is evidence that I am not really doing this work. Like my readers will see the typo and think the whole thing was generated by a machine I did not bother to check.

That feeling is real. It is also unfair. And the more I sat with it this week, the more I realized it points at something much bigger about how we are talking about AI right now.

The Coffee Shop

Let me tell you what got me thinking about this.

Imagine you own a coffee shop. You sell coffee and you sell an experience. The two are inseparable. If the coffee is bad, the experience is bad. If the employee is rude or makes a mistake on an order, the experience is bad even if the coffee is fine.

You will lose customers. You will get bad reviews. The whole business depends on getting both right, every time, for every customer.

Now think about how often that actually happens.

Coffee orders get made wrong every day in shops across the world. Employees have bad days, mishear orders, get distracted, make mistakes. Some get fixed quickly with a smile. Some leave the customer annoyed. Some end up as a one-star Yelp review.

Nobody is calling for the elimination of human baristas because of this. Nobody is writing think pieces titled "The Fundamental Flaw of Human-Powered Customer Service." We accept that humans make mistakes, that this is the cost of being served by humans, and we move on.

But when AI makes a mistake?

It is suddenly evidence of a fundamental flaw in the entire technology.

I am not saying AI hallucination is not a real problem. It is. Edition 350 ("MIT Just Taught AI to Say I'm Not Sure") covered the new MIT calibration research that is specifically designed to address this. We need AI to know what it does not know. That work matters.

But the cultural framing is something different. The cultural framing is "AI cannot be trusted because it sometimes gets things wrong."

Compared to what?

Self-Driving Cars and the New Technology Tax

Here is another version of the same problem.

Self-driving cars are now statistically safer than human drivers in most measured conditions. The data has been getting clearer for years. Per mile driven, autonomous vehicles cause fewer crashes, fewer fatalities, fewer serious injuries.

And yet, every time a self-driving car has any incident, it makes national news.

Every year, tens of thousands of people die in human-caused car accidents in the United States. We do not call for the elimination of human drivers. We accept the cost. We require licenses, we have rules, and we move on.

But a single self-driving car incident gets framed as proof that the technology is not ready.

This is the new technology tax. When something is new, we hold it to a standard we never held the thing it is replacing. We expect perfection from the new and tolerate routine failure from the familiar.

That is human nature. It is not crazy. We trust what we know. We are skeptical of what we do not.

But it does mean that a lot of arguments against AI right now are not really about AI. They are about discomfort with new technology, dressed up as concerns about reliability, accuracy, or trust.

This is happening with AI hallucination. It is happening with AI water and energy use. It is happening with AI in education. The complaints are real, but the standard being applied is one we have never applied to the alternatives.

What This Looks Like for Dyslexic Thinkers

Here is where it gets personal.

If you are a dyslexic creator using AI to help you write, code, design, or communicate, you live with this double standard every day.

When you submit a piece of writing that has small errors, it gets read as evidence of you being lazy or careless. Not as evidence that you are working with a different cognitive style. Not as evidence that proofreading is genuinely harder for your brain. Not as evidence that even with help, mistakes will sometimes get through.

When you use AI to help, and it produces something with a small error you did not catch, the framing flips. Now it is evidence that you are over-relying on technology. That you should be doing more of the work yourself. That the AI is not really helping if you cannot also be perfect.

You cannot win.

The neurotypical professional gets to make mistakes and call it being human. The AI gets framed as a fundamental flaw when it makes mistakes. The dyslexic person who uses AI to help them produce work gets accused of either being lazy (without AI) or being inauthentic (with AI).

The standards do not match. And the people getting squeezed in the middle are the ones the tools are supposed to help.

In Edition 349 ("A New Paper Just Named the Problem I Have Been Writing About for Three Years"), I responded to academic research on the LLM Fallacy by pointing out that for dyslexic creators, AI is often revealing capability that was always there, not inflating capability that was not. The wheelchair analogy. A wheelchair does not make a paraplegic falsely believe they can walk. It gives them a different way to move.

The same logic applies here. An AI that occasionally makes a mistake while helping me communicate is not making me less authentic. It is letting me communicate at all.

A Brief Note on Where This Is Headed

I want to mention something quickly because it connects.

The reason I think about this so much is that I am building toward a future where my Cognitive Partner workflows handle more of my newsletter, my content, my client work, and the rest of LM Lab AI. Agents that work in my voice. Workflows that match how I actually think. A system where I can scale my output without losing my authenticity.

If you have been here since Edition 344 ("I Woke Up at 4AM With a Random AI Idea"), you know about the Cognitive Partner OS. Edition 345 ("We Have Been Asking the Wrong Question About AI") laid out the evaluation framework that supports it.

In a near future where my newsletter is largely written through a voice-first agent that knows me well enough to draft in my style, the question of mistakes becomes more important, not less. Because if I am scaling output, even small error rates compound.

This is exactly why frameworks like the Cognitive Balance Model and the Human Guidance Index from Edition 332 matter. The human stays in the integration phase. The human catches what the AI misses. The human owns the work because the human reviewed it.

That is not a flaw in the system. That is the system working.

A baker still tastes the bread. A coach still watches the game tape. A founder still reads the contract. The human integration step is what makes co-produced work trustworthy. AI does not eliminate the need for that step. It just changes what the step looks like.

Mistakes as Features

Here is the part I cannot stop thinking about.

Humans have been embedding mistakes into our work on purpose for as long as we have been making things.

Yearbooks have always had hidden jokes and intentional mistakes that students worked to sneak past the editors. That was part of the fun. Find the typo. Find the photo that should not have made the cut. Find the inside joke buried in the captions.

Newspaper layout people have been hiding small things in their work for as long as newspapers have existed. Tiny inside jokes. Easter eggs. Letters that, when you trace them through, spell out something else.

Ancient texts have hidden meanings encoded in letter patterns, font choices, and bolded characters. People wrote things into their work that were meant to be discovered. The mistakes were sometimes the message.

I am not saying every AI typo is secretly profound. I am saying we have a long history as a species of treating imperfection as part of the work, not separate from it. The handmade pottery has the irregular glaze. The painting has visible brushstrokes. The novel has the author's stylistic quirks.

A perfect output without any human fingerprints might be exactly the kind of output we should be most skeptical of.

That is not a defense of laziness. It is a reminder that we are not actually trying to produce machine-perfect work. We are trying to produce human-meaningful work, and humans are imperfect, and that has always been okay.

OK But What Do I Actually Do With This?

Three things. The next time you find a mistake in your AI-assisted work.

1. Compare It to the Alternative

Before you spiral about an AI-related error in your work, ask: would this have happened without AI? Or would the work have not existed at all?

For a lot of dyslexic and neurodivergent creators, the honest answer is the second one. Without AI, the piece would not have been written. The error you are upset about exists inside work that exists because of the help you used. That is a different math problem than "AI ruined my output."

2. Apply the Coffee Shop Standard

If a barista misspelled your name on a cup, you would laugh, take a photo, and post it to Instagram. You would not write a 3,000-word essay about the fundamental flaws of human-powered service.

Apply that standard to AI. Catch the mistake. Fix what you can. Move on. The output is not invalidated by a small error any more than a coffee is invalidated by a misspelled name.

3. Stay in the Integration Phase

Read your work before you publish. Run it through a final pass. Use the Human Guidance Index from Edition 332. Score yourself on Human Initiation, AI Expansion, and Human Integration. Make sure your integration phase is real, not just a glance.

If you do that and a mistake still slips through, that is just being human. Welcome to the club.

What This Means for You Right Now

I want you to remember something.

The argument that AI makes mistakes is not wrong. AI does make mistakes. So do humans. The question is whether the standard we are applying is reasonable.

Right now, in April 2026, we are holding AI to a higher standard than we have ever held human work.

We tolerate routine human error in customer service, education, journalism, medicine, law, and government. We accept that humans miss things, make calls under pressure, and produce work that has flaws.

We are not extending that same grace to AI. And the people who pay the price for that double standard are not the AI companies. They are the dyslexic creators, the neurodivergent professionals, the kids with learning differences, and the small business owners who are using these tools to do work they could not do alone.

If you are one of those people, I want you to hear this clearly.

Your work is not less valid because you used AI to help. Your output is not inauthentic because a tool helped you produce it. A small mistake in your AI-assisted writing is not evidence of fraud. It is evidence that you are doing the work, in the messy, imperfect, fully human way that all good work has always been done.

Send the newsletter anyway.

Make the post anyway.

Ship the project anyway.

The coffee shop will misspell your name on the cup. The yearbook will have a typo. The self-driving car will, at some point, make a wrong call.

And the work will still be worth doing.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

☕ AI makes mistakes. So do baristas, yearbook editors, newspaper layout teams, doctors, lawyers, and self-driving cars (which crash less often per mile than humans). We extend grace for human error. We treat AI error as a fundamental flaw. The standards do not match.

🚗 This is the new technology tax. When something is new, we hold it to a higher standard than what it replaces. Most arguments against AI right now are not really about AI. They are about discomfort with new technology dressed up as concerns about reliability.

🧠 For dyslexic creators, the double standard is brutal. Without AI: lazy. With AI and a small error: inauthentic. The standard cannot be met either way. The wheelchair analogy from Edition 349 applies again.

🔧 Mistakes have always been part of human creative work. Sometimes on purpose. Yearbook pranks. Newspaper Easter eggs. Hidden meanings in ancient texts. Perfect output with no human fingerprints might be exactly what we should distrust.

⚖️ The Cognitive Balance Model from Edition 332 still applies. Human Initiation. AI Expansion. Human Integration. Stay in the integration phase. Catch what you can. Forgive yourself when something slips through.

🛠️ Three things to do when you find an AI-related mistake: compare it to the alternative (often "no work at all"), apply the coffee shop standard (laugh, fix, move on), and make sure your integration phase is real, not just a glance.

Previously

• Edition 351: "Who Is AI Actually Serving Right Now?" (Rithm survey, Gen Z sentiment, accessibility divide)

• Edition 350: "MIT Just Taught AI to Say 'I'm Not Sure'" (calibration, ternary thinking, Skeptic layer)

• Edition 349: "A New Paper Just Named the Problem I Have Been Writing About for Three Years" (LLM Fallacy, wheelchair analogy)

• Edition 345: "We Have Been Asking the Wrong Question About AI" (evaluation framework manifesto)

• Edition 344: "I Woke Up at 4AM With a Random AI Idea" (Cognitive Partner OS)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model, HGI)

• Edition 324: "When Voice Stops Working" (voice-to-text as accessibility)

Next

Edition 353: What I am actually building with LM Lab AI. The vision behind three and a half years of newsletters. A self-written newsletter. An agent for every industry I care about. A solo entrepreneur with the output of a whole company. And why dyslexic brains are the ones who see this coming first.

Edition 351 | April 28, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• What two major youth surveys just revealed about Gen Z and AI

• The strongest predictor of harmful AI use (it is not social isolation)

• Why "feeling like a burden" matters more than screen time

• How the AI accessibility divide is quietly widening, not closing

• What this means for dyslexic kids, homeschool families, and anyone fighting for cognitive equity

• Three things you can do this week regardless of your role

Reading Time: 9 minutes Listening Time: 13 minutes

Happy Tuesday.

This week I sat with two pieces of research that I cannot get out of my head.

The first is a Rithm Project survey of 2,383 young people, ages 13 to 24. They wanted to understand who is at high risk for harmful AI use. The result surprised even the researchers.

The second is a Gallup, Walton, and GSV survey of Gen Z. They asked young people how they actually feel about AI. The numbers moved in a direction nobody expected.

And then there is something a National Public Radio story flagged earlier this year that is starting to come into sharper focus: the most accurate AI tools cost the most, which means richer schools and families get better AI than poorer ones.

Put it all together, and a hard question shows up.

Who is AI actually serving right now? And who is being left behind?

Let me tell you what is really happening.

What the Rithm Survey Found

When researchers wanted to know which young people are most at risk for unhealthy AI use, they expected the answer to be obvious. Lonely kids. Isolated kids. Kids without friend groups.

That is not what they found.

The strongest predictor of high-risk AI use was young people feeling like a burden or being unable to be authentic with other people. Students with friend groups could still be at risk. Kids surrounded by people who loved them could still slide into dependence on AI for emotional connection.

The trigger was not loneliness. It was the feeling of not being able to be yourself with the people around you.

I want you to sit with that for a second.

For dyslexic and neurodivergent kids, this is a punch in the chest. Because that feeling, the feeling of not being able to be authentic, of sensing you are too much or too different or too slow or too weird for the room, is something many of us carried for our entire childhoods.

It was the assumption behind every red pen mark on a spelling test. It was the silence after I read out loud in class. It was every time someone said "just try harder" when trying harder was already the only thing I had.

And the survey is telling us, with data, that kids who feel that way are now turning to AI in ways that hurt them.

Not because they are weak. Not because they lack discipline. Because they finally found something that listens without judgment, and they are spending too much of themselves on it because the alternative feels worse.

The Gen Z Sentiment Crash

If the Rithm survey told us who is at risk, the Gallup survey tells us how the broader generation feels.

Excitement about AI dropped 14 points in a single year. Only 22 percent of Gen Z respondents now describe themselves as excited about AI. Only 18 percent feel hopeful, down 9 points. Anger rose 9 points.

This is striking because Gen Z is not a generation that is anti-technology. They grew up with smartphones. They are the most digitally fluent generation in history. And they are losing patience with AI faster than any other group.

Why?

I have a theory. It is not just about AI being scary or hyped or unreliable. It is about something more specific.

Young people are not seeing themselves in the version of AI that is being sold to them.

Most AI tools are still optimized for office productivity, professional writing, and tasks that look like adult work. The flashy demos are about coding, business analysis, and corporate workflows. The voices in the AI conversation are mostly venture capitalists, tech executives, and engineers.

Where is the AI that helps a 14-year-old dyslexic kid finally read a chapter book?

Where is the AI that gives a homeschool parent an evaluation framework for choosing tools that fit their kid?

Where is the AI built around the question "what does this person actually need" instead of "what can we sell to enterprises"?

That gap is part of what Gen Z is reacting to. And honestly, I do not blame them.

The Accessibility Divide Nobody Wants to Talk About

Earlier this year, a Brookings Institution report covered by NPR raised something that should bother all of us.

The report, by global education expert Rebecca Winthrop, pointed out that AI can absolutely make classrooms more accessible for students with learning disabilities, including dyslexia. That part is true.

But Winthrop also flagged something most coverage skipped over.

"AI can massively increase existing divides."

Why? Because the free AI tools that are most accessible to students and schools are also the least reliable and least factually accurate. The most accurate, most useful AI models cost more. Which means richer communities and richer schools can afford better AI.

For the first time in education technology history, schools have to pay more for more accurate information.

Read that again. The kids who already have the most resources will get the most accurate AI. The kids who already have the least will get the noisiest, least reliable AI.

The "AI revolution" for education risks becoming another way the gap gets wider.

This is not theoretical. This is happening now. Edition 343 ("Stanford Just Measured Everything") showed that AI capability is accelerating faster than any technology in history. Edition 339 ("Your AI Just Forgot Everything") showed that the most useful AI is the AI with the best memory and context, which is not the free version.

The kids who need AI accessibility the most, dyslexic kids, neurodivergent learners, kids in under-resourced schools, are the most likely to get the version of AI that fails them.

That is not progress. That is automated inequality.

Why These Three Findings Belong Together

I want to draw the threads together.

Finding 1: The strongest predictor of harmful AI use among young people is feeling like they cannot be authentic with the people around them.

Finding 2: Gen Z is increasingly frustrated and angry about AI, with positive sentiment cratering in a single year.

Finding 3: The most accurate AI is the most expensive AI, and the kids who most need accurate AI are the least likely to get it.

What is the connection?

The kids who would benefit most from genuinely good AI are the kids being given the worst version of it.

A dyslexic kid in a wealthy private school district might get access to Claude Pro or ChatGPT Plus, with proper memory features, design taste, and accuracy. They get a tool that reduces cognitive load.

A dyslexic kid in an underfunded public school gets a free chatbot with limited memory, fewer features, and lower accuracy. They get a tool that adds cognitive load while pretending to reduce it.

Both kids feel like burdens. Both kids are looking for something that lets them be authentic. One gets a real tool. The other gets a hollow version that may eventually fail them.

This is what the Rithm survey is picking up on. The kids who feel they cannot be authentic are turning to AI. And the AI they can afford might be exactly the wrong AI.

In Edition 342 ("The Weight in My Chest"), I said I refuse to wait for someone else to tell me how AI changes my life. In Edition 345 ("We Have Been Asking the Wrong Question About AI"), I made the case that evaluation is the most important skill of the AI era.

These two ideas land differently when you realize that most of the kids who need this work the most cannot do it for themselves.

That is on us.

What I Have Been Building Toward

This is exactly why the work I have been doing matters more than I sometimes realize.

The Cognitive Partner OS from Edition 344. The evaluation frameworks from Edition 345. The Homeschool Parent's Guide to AI from Edition 337. The family evaluation tool I started building in Edition 341. The cognitive fit principle. The Single Source of Truth. The Cognitive Balance Model.

Every single one of these tools and frameworks exists because the default version of AI does not fit dyslexic and neurodivergent users. It does not fit families. It does not fit kids. It does not fit anyone outside the wealthy professional knowledge worker who is the implicit default user that AI products are designed for.

If we do not build for the kids in the Rithm survey, nobody else will.

If we do not build for the families who cannot afford Claude Pro or ChatGPT Plus, nobody else will.

If we do not push for accessibility-first, neurodivergent-first, family-first design, the divide that NPR flagged is going to become permanent.

This is the work. And I want you to know that when I write about evaluation frameworks and prompt libraries and homeschool agents, I am thinking about that 14-year-old dyslexic kid who is using a cheap AI chatbot to do her homework because she cannot read the textbook and her school cannot afford the better tool.

She is the reason any of this matters.

OK But What Do I Actually Do With This?

Three things. This week. No matter what role you have.

1. If You Are a Parent or Caregiver

Have one honest conversation with the young person in your life about how they actually use AI. Not what they tell you they use it for. What they really do with it.

Ask: "What do you talk to AI about that you do not feel like you can talk to me about?" The answer might break your heart. It might also be the most important thing you hear this year.

The Rithm research says human connection protects. Connection does not mean control. It means being someone they can be authentic with.

2. If You Work in Education or Have Influence in a School

Look at what AI tools your students actually have access to. Is your school giving them the free, less accurate version of AI? Are you putting better tools in front of the kids who need accommodations the most, or are those kids getting the worst tools?

This is a budget question. It is also a values question. Push for the conversation.

3. If You Are a Builder, Founder, or Educator With Reach

Build for the kids who do not have access. Build for the families that cannot pay for premium tools. Make your evaluation frameworks free. Make your prompt libraries free. Make the version of AI that actually fits a dyslexic brain available to the kid who needs it most, not just the kid whose parents can afford it.

If you are a Cognitive Partner Member, this is your community too. Help me figure out how to get this work into the hands of the people who need it without paywalls in the way.

What This Means for You Right Now

Here is the honest version.

AI is not going to revolutionize education on its own. It is not going to lift up neurodivergent kids by itself. It won't make the world more accessible by itself.

It will go where the money goes. That is what every previous educational technology has done. That is what current AI development is already showing us. The most resourced communities get the best tools. The least resourced get the leftovers.

Unless people like us push back. Unless we build deliberately for the kids being left behind. Unless we make the case, with our own work, that the future of AI has to include the people the present version is failing.

The Rithm survey told us what kids actually need: to feel seen, safe, and free to be authentic. The Gallup survey told us they are losing faith. The accessibility report told us the divide is getting wider, not narrower.

These are not three separate problems. They are one problem with three faces.

The solution is not more AI. The solution is better AI, applied with intention, by people who refuse to let it become another tool of inequity.

That work starts wherever you are. With whoever you can reach. With whatever tools you can afford or create.

If you have made it this far in this newsletter, you are already part of it.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

📊 A new Rithm Project survey of 2,383 young people found the strongest predictor of harmful AI use is feeling like a burden or being unable to be authentic with others. Not social isolation. Not screen time. Authenticity.

📉 A Gallup, Walton, and GSV survey of Gen Z shows AI excitement dropped 14 points in a single year. Anger rose 9 points. Only 22% are excited. Only 18% are hopeful.

⚖️ A Brookings/NPR analysis flagged that the most accurate AI tools cost the most. For the first time in education technology history, schools must pay more for better information. The accessibility divide is widening, not closing.

🧠 Dyslexic and neurodivergent kids are exactly the kids who need accurate, well-designed AI the most. They are also the kids most likely to get the worst version of AI.

🏗️ Every framework I have built (Cognitive Balance Model, HGI, Single Source of Truth, evaluation tools, Cognitive Partner OS) exists because the default version of AI does not fit the people who need it most. This is the work.

🛠️ Three things to do this week: have an honest conversation with a young person about how they actually use AI, audit what tools your school or family actually has access to, and build deliberately for the people being left behind.

Previously

• Edition 350: "MIT Just Taught AI to Say 'I'm Not Sure'" (calibration, ternary thinking, the Skeptic layer)

• Edition 349: "A New Paper Just Named the Problem I Have Been Writing About for Three Years" (LLM Fallacy, Cognitive Balance Model)

• Edition 345: "We Have Been Asking the Wrong Question About AI" (evaluation framework manifesto)

• Edition 344: "I Woke Up at 4AM With a Random AI Idea" (Cognitive Partner OS)

• Edition 343: "Stanford Just Measured Everything About AI" (AI Index, jagged frontier)

• Edition 342: "The Weight in My Chest" (autonomy, sticktoitness, sovereignty)

• Edition 341: "I Have Never Seen Anything Like This Before" (state of AI, evaluation tools started)

• Edition 339: "Your AI Just Forgot Everything. Again." (memory architecture)

• Edition 337: "I Gave My Old Mac Mini a Brain" (Homeschool Parent's Guide to AI preview)

• Edition 325: "My 14-Year-Old Daughter Just Proved Me Wrong" (Makena, family AI)

Next

Edition 352: We hold AI to a standard we have never held ourselves to. Coffee orders get made wrong. Yearbooks have typos. Self-driving cars crash less than humans. Why is "AI makes mistakes" the argument that wins? A look at the double standard, why it hurts dyslexic creators most, and the personal worry behind the whole thing.

Edition 350 | April 23, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• What MIT just figured out about AI confidence and why it matters

• The training problem that has been quietly making AI models worse

• Why the "third option" (maybe, I do not know, I am not sure) is more powerful than people realize

• How ternary thinking connects to dyslexic and neurodivergent cognition

• Why this might be the missing piece in cognitive architecture research

• Three things you can do today to bring this calibration into your own AI use

Reading Time: 9 minutes Listening Time: 12 minutes

Happy Monday,

May the 4th be with you 😉

A paper dropped out of MIT's CSAIL lab last week that I cannot stop thinking about.

The headline: researchers taught AI to say "I am not sure."

That sounds simple. It is not. It is one of the most important findings in AI training I have seen this year. And it connects to something I have been writing about, in different forms, for longer than I realized.

Let me explain.

The Problem MIT Just Named

Today's reasoning AI models have a confidence problem.

When you ask one of them a question, it answers with the same level of certainty whether it actually knows the answer or it is essentially flipping a coin. The most powerful systems in the world will tell you something with complete confidence and be completely wrong, and you have no signal to know the difference.

MIT researchers traced this back to a specific flaw in how the models are trained.

The standard approach: reward the model for getting the right answer. Penalize it for getting the wrong answer. Nothing in between.

The result: the model learns to confidently produce an answer no matter what. There is no incentive to express uncertainty. There is no reward for saying "I am not sure." The training process literally teaches confident guessing.

In MIT's words, "The standard training approach is simple and powerful, but it gives the model no incentive to express uncertainty or say I do not know. So the model naturally learns to guess when it is unsure."

That is a problem with consequences.

In medicine, finance, law, or any setting where people make real decisions based on AI output, a system that says "I am 95 percent sure" when it is actually right half the time is more dangerous than a system that gets the answer wrong outright. Because there is no warning signal.

You can spot a wrong answer. You cannot easily spot a wrong confidence level.

The MIT Fix

Here is what they did.

They built a method called RLCR, which stands for Reinforcement Learning with Calibration Rewards. Instead of just training models to be right or wrong, they added a single new term to the reward function: a measure that penalizes the gap between how confident the model says it is and how often it is actually correct.

Now the model is trained to do two things at once. Solve the problem. And accurately estimate its own uncertainty.

The results were striking. RLCR reduced calibration error by up to 90 percent without losing accuracy. Both on tasks the model was trained on and on entirely new ones it had never seen.

Even more interesting: standard reinforcement learning training was actively making models worse at calibration. Not just failing to help. Actually hurting it. The MIT researchers said it directly: "The models become more capable and more overconfident at the same time."

That is the part I want you to sit with for a second.

The most advanced AI training methods in the world were making models simultaneously smarter and worse at knowing what they did not know.

This is the issue our entire industry has been ignoring while sprinting toward bigger models and faster benchmarks.

The Third Option

Here is where this gets personal for me.

Years ago, I wrote about ternary systems in this newsletter. The idea that human thinking is rarely a clean yes-or-no proposition. There is almost always a third option.

Yes. No. Maybe.

That third one is where most of the real cognitive work happens. Maybe is not weakness. Maybe is honesty. Maybe is the space where you slow down, gather more information, ask another question, hold uncertainty without pretending you have resolved it.

For most of computing history, we have built systems that ignore the third option. Binary logic. True or false. One or zero. Even most AI training, until this MIT work, only rewarded the binary outcomes.

And here is where it lands for me, even if I cannot prove the connection scientifically.

My brain has always lived in the third option.

A lot of dyslexic and neurodivergent thinkers will recognize this. We are not always rapid yes-or-no machines. We sit with information. We hold things in tension. We see multiple possibilities at once and we are slow to collapse them into a single answer.

For most of my life, that was framed as a deficit. Indecisive. Slow to commit. Unfocused.

But what if it was actually calibration?

What if the brains that naturally live in the "I am not sure" space have been doing exactly what MIT is now teaching AI to do?

I cannot prove this is real correlation. I am not a researcher. I am a guy with a recreation degree and a lot of time spent thinking about how my brain actually works.

But the more I learn about cognitive architecture, the more I think the brains that hold uncertainty well are not broken. They are doing something the rest of the system has been undervaluing.

Why This Connects to Cognitive Architecture

In Edition 339 ("Your AI Just Forgot Everything. Again."), we walked through the five-layer self-improving AI stack from Chappy Asel's research. Memory. Knowledge. Context. Skills. Self-improvement.

In Edition 343 ("Stanford Just Measured Everything"), we looked at the AI Index and the "jagged frontier" finding. Capability is spiky. Brilliant in some areas, weak in others.

This MIT work fits underneath all of that.

Because if a model cannot accurately report its own uncertainty, every layer above it gets compromised. Memory gets contaminated with false-confidence facts. Knowledge bases get filled with overconfident assertions. Self-improvement loops get trained on flawed signals.

Calibrated uncertainty is the foundation that everything else sits on.

And it is the foundation that has been quietly missing.

The same is true for human cognitive architecture. We do not have a complete map of how human memory, reasoning, and uncertainty work together. We know there is short-term and long-term memory. We know there are different processing styles. We know dyslexic brains process visually and spatially in ways that neurotypical brains often do not.

What we do not have is a clean understanding of how uncertainty itself is processed. And whether brains that hold uncertainty differently are operating in a less effective mode, or just a different one.

The MIT paper does not answer that question. But it points at something important. Uncertainty is not a bug. It is signal. A model that knows what it does not know is more useful than a model that guesses with confidence.

That principle, applied to humans, suggests the people who naturally hold uncertainty without rushing to resolve it might be more useful in certain domains than the people who resolve everything quickly.

That is a different way to think about neurodivergent cognition. And it is one I am going to keep pulling on.

Where This Connects to Frameworks I Have Already Built

This is the part I want to highlight because it surprised me when I noticed it.

I have been building a three-layer cognitive partner architecture for a while now. The DLM Three-Layer Architecture: Socratic, Strategic, Skeptic.

The Socratic layer asks better questions.

The Strategic layer makes better plans.

The Skeptic layer holds uncertainty. It says "wait." It says "we do not know yet." It says "let me check that before we commit."

That third layer has always been important to me, but I am realizing now that I built it for a reason I could not fully articulate at the time. I built it because every reasoning system needs a third option. Every cognitive architecture needs a place where uncertainty lives without being punished.

MIT just gave me the language for what I was building toward.

The Skeptic layer is the calibration layer. It is the place where "I am not sure" is allowed to exist. It is the part of the architecture that prevents confident wrong answers from cascading into worse decisions.

Three years of newsletter work, and I am only just now seeing how this one piece fits.

That is the part of building in public that keeps me coming back. Sometimes you do not understand what you are building until someone else's research names it for you.

The Cognitive Balance Model Connection

This also lines up with the Cognitive Balance Model from Edition 332 ("A Year Ago, I Was in a Hospital Bed").

The model has three phases: Human Initiation, AI Expansion, Human Integration. Scored on the Human Guidance Index from 3 to 15.

But underneath all three phases, there is an implicit fourth thing. Honest uncertainty.

Human Initiation is stronger when the human is honest about what they do not know yet. AI Expansion is more useful when the AI can say "I am not confident in this part." Human Integration is more reliable when both sides can flag uncertainty for review.

Calibration is not a separate phase. It is a quality that runs through all three phases.

A high HGI score requires honest uncertainty at every level. Without it, the score is just confidence in motion. With it, the score reflects real collaboration.

I think this is going to become an explicit part of how I teach the framework going forward. Not just the three phases. The calibration that runs through them.

OK But What Do I Actually Do With This?

Three things. This week.

1. Ask Your AI to Show Its Confidence

The next time you ask Claude or ChatGPT or any AI for something important, add this to the end of your prompt:

Most current models do not do this naturally. But they will if you ask. And what comes back is often more useful than the answer itself.

This is a mini version of what RLCR is trying to teach. You are forcing the calibration manually because the training has not done it yet.

2. Practice the "Maybe" in Your Own Thinking

Notice when you are pressured to give a yes-or-no answer that does not have one. Notice the weight of that pressure. Notice the moments where "I am not sure" would actually be the most accurate response.

Practice saying it out loud. To yourself. To your team. To your AI.

For a lot of dyslexic and neurodivergent thinkers, this comes naturally and gets trained out of us by environments that do not reward it. Reclaiming that third option is part of working with how your brain actually wants to operate.

3. Add a Skeptic Layer to Your Single Source of Truth

If you have built a Single Source of Truth from Edition 329, add a section called something like "What I Am Uncertain About Right Now."

List the strategic decisions you are sitting with. The questions you have not answered. The directions you have not picked. Update it regularly.

This becomes the calibration layer of your own cognitive partner setup. Your AI gets to see not just what you know, but where you are still working things out. That changes the quality of the collaboration in ways that surprised me when I started doing it.

What This Means for You Right Now

We are at a moment in AI history where the most important findings are not always about new capabilities. Sometimes they are about acknowledging what the capabilities are missing.

MIT just acknowledged that the most advanced reasoning AI models in the world have been trained to be confidently wrong. And they fixed it with a single change to the reward function. A 90 percent reduction in calibration error without losing accuracy.

That is huge.

For our community, the lesson is bigger than the technical fix.

Uncertainty is signal, not weakness.

For the brains that have always lived in the third option, that is validation. Not just emotional validation. Architectural validation. The cognitive style that has been undervalued is the same style that AI systems are now being deliberately trained toward.

We have been doing this all along. We were not broken. We were calibrated.

And the world is starting to catch up.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously. 

TL;DR- For My Fellow Skimmers

🧠 MIT's CSAIL lab just published research showing that the most advanced AI training methods have been making models simultaneously smarter and worse at knowing what they do not know.

🎯 Their fix: RLCR (Reinforcement Learning with Calibration Rewards) trains models to estimate uncertainty alongside their answers. It reduced calibration error by up to 90% without losing accuracy.

⚖️ The deeper insight: a model that says "I am 95% sure" when it is right half the time is more dangerous than one that gets the answer wrong outright. Calibration is the foundation everything else sits on.

🔀 Ternary thinking matters. Yes, no, maybe. The third option is where most real cognitive work happens. Most computing history has ignored it.

🧩 For dyslexic and neurodivergent thinkers: many of us have always lived in the "I am not sure" space. What gets framed as indecisive may actually be calibration. The cognitive style being undervalued is the one AI is now trained toward.

🏗️ This connects to the DLM Three-Layer Architecture (Socratic, Strategic, Skeptic) and the Cognitive Balance Model. The Skeptic layer is the calibration layer. Honest uncertainty runs through all three phases of the HGI.

🛠️ Three things to do this week: ask your AI to show its confidence, practice the "maybe" in your own thinking, and add a "what I am uncertain about" section to your Single Source of Truth.

Edition 349 | April 22, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• What a new academic paper is calling the "LLM Fallacy"

• Why researchers are finally noticing what we have been living with

• Why the paper is correct about the risk and partially wrong about the solution

• Why the Cognitive Balance Model is the antidote

• How the HGI (Human Guidance Index) gives you a measurable way to avoid this trap

• Three honest questions to ask yourself about your own AI work this week

Reading Time: 9 minutes Listening Time: 13 minutes

Happy Wednesday.

A paper dropped on arXiv last week that I have been sitting with for a few days.

It is called "The LLM Fallacy: Misattribution in AI-Assisted Cognitive Workflows", by researchers at ddai Inc. And it names something I have been writing about for three and a half years, but from an angle I want to push back on.

Here is the paper's core argument, in plain language.

When you use a large language model to help you write, code, analyze, or communicate, the output often feels like your own work. The interaction is smooth. The model disappears into the background. You end up with something that looks and sounds like you produced it yourself.

The researchers argue that this creates a cognitive attribution error. You start to misinterpret AI-assisted outputs as evidence of your own independent competence. Over time, a gap opens between what you think you can do and what you can actually do without the tool.

They call this the LLM Fallacy.

And they are right about the risk.

But I think they are partially wrong about the solution. And I want to tell you why.

What the Paper Gets Right

Let me steelman it first.

The paper makes a real observation. When the tool is fluent, fast, and invisible enough, it becomes very easy to confuse "what we did together" with "what I did alone."

That is not just a theoretical concern. It is a practical one. If a student uses an LLM to write an essay and gets an A, they may walk away thinking they have the ability to write that essay. If a junior engineer ships code that Claude wrote, they may believe their technical skills are growing faster than they are.

This matters for education. For hiring. For skill development. For self-knowledge.

The paper situates this inside existing research on automation bias, cognitive offloading, and the Dunning-Kruger effect. It argues that LLM workflows create a new kind of attributional distortion because of how opaque, fluent, and low-friction they are.

I agree with all of that.

I have been warning about it in different words since Edition 329 ("Building Your Second Brain"), where I introduced the Single Source of Truth framework specifically so your AI would reflect you back rather than overwrite you.

The risk is real.

Where the Paper Stops Short

Here is what the paper does not say.

It treats the LLM Fallacy as a general human problem with AI-assisted workflows. Opacity plus fluency plus low friction equals attribution error. That formula applies to everyone, in theory.

In practice, it lands very differently on different brains.

For a neurotypical professional who has always been able to write, code, or analyze competently on their own, an LLM can absolutely create the illusion of skills they do not actually have. The paper is writing for that person.

For a dyslexic thinker who has spent a lifetime being told that they are less capable because they struggle with spelling, sequential text, or conventional writing formats, an LLM is doing something completely different.

It is not inflating a competence that does not exist.

It is exposing a competence that was always there and could never get through.

That is a critical distinction that this paper misses entirely.

In Edition 324 ("When Voice Stops Working"), I argued that voice-to-text is not a convenience for me. It is an accessibility tool. It is the bridge between the ideas in my head and the words on a screen. Before these tools existed, the bridge was broken. My ideas never made it across at full resolution. That is not a fallacy. That is a disability being accommodated.

A wheelchair does not make a paraplegic "falsely believe" they can walk. It gives them a different way to move. The goal was never walking. The goal was movement.

That is what AI is for a lot of dyslexic thinkers. The goal was never writing in the conventional sense. The goal was communicating. And now we can.

But the Risk Is Still Real

I do not want to wave away the paper's warning just because the framing is incomplete.

Because here is the thing. Even for dyslexic creators using AI as accessibility, the fallacy can still show up. Just in a different shape.

A dyslexic entrepreneur might believe they are building a scalable business on AI output when in reality they have not developed the underlying strategic thinking.

A neurodivergent student might believe they understand a subject because they can produce polished essays about it with AI assistance when in reality they have not done the learning.

A homeschooling parent might believe their child is progressing academically when in reality the AI is doing most of the work the child should be doing.

All of these are real. None of them are imaginary.

The LLM Fallacy is not just a risk for neurotypical professionals. It is a risk for anyone who stops paying attention to where the work is actually coming from.

That is the real question. Not "is the tool helping?" Of course it is.

The real question is: "How much of this is me, and how much of this is the tool, and do I know the difference?"

If you do not know the difference, the fallacy has you.

If you do know the difference, you are in a completely different position.

The Cognitive Balance Model as the Antidote

Here is where I want to push this conversation forward.

Back in Edition 332 ("A Year Ago, I Was in a Hospital Bed"), I introduced the Cognitive Balance Model. It is a framework for how humans and AI should work together. Three phases:

Human Initiation. You set the direction. You decide what problem to solve, what quality bar to hit, what outcome matters.

AI Expansion. The AI does the heavy lifting. Drafts. Alternatives. Calculations. Research. Options.

Human Integration. You make the final call. You review, edit, approve, or reject. The output is yours because you decided what passes.

Each phase gets a score from 1 to 5 on the Human Guidance Index (HGI). Total score from 3 to 15. The higher the score, the more human guidance is in the loop at every phase.

Here is what the LLM Fallacy paper does not know.

The fallacy only happens when HGI is low.

If you type a vague prompt, accept the first output, and paste it somewhere, your HGI is around 3 or 4. Human Initiation is weak. AI Expansion is the whole show. Human Integration is basically zero.

That is where the attribution error lives. That is where the illusion of skill takes hold. That is where you lose track of where the thinking came from.

But if you enter the interaction with a clear intention, give the AI specific context, push back on outputs that miss, iterate deliberately, and make integration calls based on your own judgment, your HGI is 12, 13, 14 out of 15.

At that level, the output is genuinely co-produced. You know what you contributed. You know what the tool contributed. There is no fallacy because there is no misattribution.

The Cognitive Balance Model does not prevent AI collaboration. It preserves the boundary that the paper is worried about eroding.

And it does it in a way that works whether you are a neurotypical professional or a dyslexic creator. The framework does not care about your cognitive style. It cares about the structure of the collaboration.

The HGI as a Diagnostic

I want to give you something practical.

The next time you finish a piece of work that AI helped you with, run this quick check:

Human Initiation (1-5): How specific was my starting point? Did I know what I wanted, or did I let the AI guide me into an answer?

AI Expansion (1-5): Did I use the AI to explore options, or did I take the first thing it gave me?

Human Integration (1-5): Did I meaningfully change, improve, or challenge the output, or did I just clean it up and move on?

If your score is below 9, you are in fallacy territory. Not because the work is bad. Because you may not know how much of it is yours.

If your score is 9 to 12, you are collaborating. Good territory.

If your score is 13 to 15, you are using AI as a genuine cognitive partner. You know exactly what you brought. You know exactly what the tool brought. You built something together that neither of you would have built alone.

That is the model. That is the antidote.

The paper identifies a real problem. The HGI gives you a way to measure whether you are walking into it or walking around it.

The Dyslexic Dimension They Missed

One more thing, because I cannot write this edition without saying it.

The paper talks about the LLM Fallacy as if it applies uniformly to all knowledge workers. It does not.

For dyslexic and neurodivergent creators, there is a separate phenomenon happening alongside the fallacy the paper describes. I do not have a perfect name for it yet, but it goes something like this.

Before AI, I was underproducing. My actual capability was high but my output was low because the conventional tools were broken for my brain.

With AI, my output has caught up to my capability. Not because my capability grew. Because the tool finally reads in the language I have always been speaking.

The LLM Fallacy paper assumes that AI output is inflating users above their actual capability. For many neurodivergent users, AI output is simply revealing capability that was always there.

These are not the same thing. And any honest framework has to hold both truths at once.

• Some users are being lifted above their skill level by the tool.

• Some users are being unblocked from an expression barrier the tool removes.

You can be in both categories at once. A dyslexic entrepreneur can have genuine strategic insight that AI finally lets them articulate (real capability, newly unblocked) AND overestimate their technical depth because AI wrote the code (fallacy, real risk).

The Cognitive Balance Model handles both. Because HGI does not measure whether the tool helped. It measures whether you stayed in the loop at every phase.

If you stayed in the loop, both things are true and manageable. If you did not, only the fallacy wins.

OK But What Do I Actually Do With This?

Three things.

1. Score Your Last Three Pieces of AI-Assisted Work

Pull up the last three things you did with AI help. Run each one through the HGI. Human Initiation, AI Expansion, Human Integration, 1 to 5 each. Honest scores.

You will probably find that some of your work was genuine collaboration and some of it was drift. That is useful information.

2. Ask Yourself the Hard Attribution Question

For one piece of recent AI-assisted work, ask: "If I had to reproduce this without AI right now, could I?"

This is not about whether you should have to. You usually should not have to. The question is whether you still know where your actual competence ends and the tool's starts.

If you can reproduce the thinking (even if it would take you longer), you own the work. If you cannot, you have attribution homework to do.

3. Raise Your HGI Deliberately on Your Next Project

Pick your next AI-assisted project and set a minimum HGI of 12 out of 15. That means stronger initiation, more deliberate iteration, more meaningful integration. It will take a little more time. You will learn more from the work. And you will own the output in a way that no paper on the LLM Fallacy can dispute.

What This Means for You Right Now

Researchers are finally starting to ask the questions this newsletter has been asking for years.

The LLM Fallacy is real. The risk is real. The attribution error is real.

And the answer is not to use AI less. The answer is to use AI with more structure, more intention, and more measurement.

That is exactly what the Cognitive Balance Model was built for. That is exactly what the Human Guidance Index measures. That is why I have been writing about this since Edition 332.

The academic world is starting to name the problem. Our community already has the framework. That is a lead, and we should use it.

For dyslexic and neurodivergent creators specifically: you are not walking into the LLM Fallacy every time you use AI. In many cases, you are walking out of a lifetime of being underestimated. The tools that finally fit your brain are revealing capability that was always there.

But you still need the framework to stay on the right side of that line. Not because you are suspect. Because the tools are powerful and the stakes are real.

Know where you are. Score your work. Stay in the loop at every phase.

That is how we use AI without losing ourselves to it.

Previously

• Edition 348: Claude Design follow-up with a real brief and actual results

• Edition 347: "I Played With Claude Design for Ten Minutes" (Anthropic Labs launch, design accessibility for neurodivergent creators)

• Edition 346: "The Meta Layer" (builder evaluation, cognitive fit)

• Edition 345: "We Have Been Asking the Wrong Question About AI" (evaluation framework manifesto)

• Edition 343: "Stanford Just Measured Everything About AI" (AI Index 2026, jagged frontier)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model and HGI introduction)

• Edition 329: "Building Your Second Brain" (Single Source of Truth)

• Edition 324: "When Voice Stops Working" (voice-to-text as accessibility)

Next

Edition 350: MIT just taught AI to say "I am not sure." Why this calibration breakthrough matters more than people are noticing, why ternary thinking (yes, no, maybe) might be the most important cognitive architecture insight of the year, and why dyslexic and neurodivergent brains may have been doing this naturally all along.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

📄 A new paper names the "LLM Fallacy": users misattributing AI-assisted output as evidence of their own independent competence. It is a real phenomenon. The risk is real.

🎯 The paper frames this as a general human problem. In practice, it lands very differently on neurodivergent brains. For dyslexic creators, AI often reveals capability that was always there, not inflates competence that was not.

⚖️ The Cognitive Balance Model is the antidote. Three phases: Human Initiation, AI Expansion, Human Integration. Scored on the HGI from 3 to 15.

📊 If your HGI is below 9, you are in fallacy territory. If it is 13 to 15, you are using AI as a genuine cognitive partner with clear attribution.

🔍 Two things can be true at once: AI can unblock real capability in neurodivergent users AND create attribution errors in the same users on different tasks. The framework handles both.

🧩 Three things to do this week: score your last three AI-assisted pieces on the HGI, ask if you could reproduce them without AI, and set a minimum HGI of 12 for your next project.

🔒 The academic world is naming problems we already have frameworks for. Cognitive Partner Members get those frameworks first. 50 founding spots at $19/month, locked forever.

Edition 348 | April 21, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• What happened when I actually gave Claude Design a proper brief

• Why I ran out of credits and what that taught me about pricing for experimentation

• The distinct visual style that Claude Design seems to lean toward

• Why the prompt had a head start and why that matters

• What this all means for my Cognitive Partner OS build

• Two things you can do this week to test Claude Design for yourself

Reading Time: 7 minutes Listening Time: 10 minutes

Happy Tuesday.

In Edition 347, I told you I was going to come back with a real experiment. Give Claude Design a proper brief. Build a real Cognitive Partner OS mockup. Share what happened honestly.

Here is what actually happened.

I gave it a very long, very specific prompt to recreate something I had already built in another project. I wanted to see how it would handle a real brief, not a throwaway one. And I wanted to compare its output to what I have been getting from other vibe coded tools.

Then I ran out of credits.

It did not even finish what I asked it to do.

But what it created before it stopped was enough to tell me something important.

Let Me Give You the Honest Picture

First, the limits.

This was not a clean, isolated experiment. The prompt I used had a head start. Because I was asking Claude Design to recreate something I have already iterated on heavily in another vibe coding tool. So it was working with a clear, mature, well-tested idea. Not starting from scratch.

That is relevant. Because part of why the output was as good as it was, I think, is that the brief was refined. Multiple versions of this project already exist. The requirements are clean. The goals are clear.

A great prompt, built from months of iteration, is going to get a better result from any tool than a quick ad-hoc ask would. That is not news. But it is worth naming here, because I do not want to tell you "Claude Design is amazing" without also telling you what I gave it to work with.

Second, the credits.

I burned through my Claude credits faster than I expected, and the system stopped the build mid-project. I will not have more credits to keep testing until next week. So this edition is built on partial output, not a finished deliverable.

That said, partial output was enough.

What I Actually Got

What Claude Design produced before I hit the credit wall was, frankly, some of the best visual output I have seen from any vibe coded tool.

And I have tried a lot of them. Edition 346 walked through the six builders I have been evaluating for the Cognitive Partner OS: Claude Code, Cursor, Codex, Google AI Studio, Bolt.new, Lovable. Each one has strengths. Each one has a look and feel.

Claude Design is different. And not just in a "they all look slightly different" way. In a noticeable, has-its-own-identity way.

It has an aesthetic. A real one.

I noticed the color choices immediately. Softer tones. Pastel accents. Light, airy backgrounds. The kind of palette that feels like it came out of the same visual design language as Claude's own interface and marketing. There is a distinct Claude look.

Fonts leaned cleaner and more readable than what I usually get from Claude Code directly. Proportions felt considered. Hierarchy was present without me having to fight for it.

It looked like a designer had been in the room.

That is a big deal for someone like me. I do not want to become a designer. I want to ship products that do not embarrass me visually. And for ten minutes of work with a detailed prompt, what I got was already ahead of what most dyslexic founders I know can produce with traditional tools in an afternoon.

The Claude Aesthetic

Let me dwell on this for a second because I think it matters.

Most vibe coding tools are neutral about visual style. They generate whatever you describe. If you do not describe it well, you get something generic. If you describe it carefully, you get something reasonable.

Claude Design seems to have a default taste.

Not overbearing. Not locking you into one look. But a baseline that tends toward calm, clean, legible, softer. Claude's own aesthetic, applied to whatever you are building.

I mentioned in Edition 347 that Claude Opus 4.7 (the model behind Claude Design) has developed something close to design judgment. Anthropic's own launch quotes talked about it making choices "I would actually ship."

After using it with a real brief, I believe it.

This does not mean every project should look like Claude. It means that the tool has a center of gravity. And for someone like me, who struggles to articulate visual preferences in precise design language, having a tool with its own tasteful default is genuinely useful.

I can start from its aesthetic and push outward, rather than starting from blank and trying to describe what "good" means.

That is a different kind of collaboration. And it fits the Cognitive Balance Model I wrote about in Edition 332 in an interesting way. The tool is contributing more than execution. It is contributing taste. I am contributing direction, context, and judgment about fit. We are building something together that neither of us would have produced alone.

What This Tells Me About the Stack

Going back to the builder evaluation from Edition 346, this experiment gave me a clearer picture of where Claude Design fits.

It is not replacing Claude Code for me. Claude Code stays the backbone for complex logic, long-running builds, and iterative development.

It is not replacing Bolt.new either. Bolt is where most of my current websites live, and it has its own strengths for fast site builds.

Claude Design is the visual and prototyping layer sitting on top. The place I go when:

• I need a dashboard or UI concept that looks professional

• I need a one-pager or pitch asset fast

• I want to see what a product idea looks like before committing to a full build

• I want a starting point with actual design taste baked in

And because it hands off to Claude Code cleanly, the workflow I am starting to see is: design the vision in Claude Design, then build the engine in Claude Code, then polish the front-end in Bolt or Lovable if needed.

That is a stack. A real one. And it is starting to feel like a stack that actually fits the way my brain works.

The Credit Question

One more honest note.

Running out of credits mid-build was frustrating. It is also a real part of the AI builder experience right now. Every one of these tools costs something. Claude Design is part of the Pro, Max, Team, or Enterprise Claude plans, but it uses credits the way other advanced features do.

For a solo builder iterating heavily, this matters.

In Edition 345, I argued that evaluation is the most important skill of the AI era. This is part of that. Understanding not just what a tool can do, but what it costs you to actually use it for real work.

I did not find out where the ceiling was on my plan until I hit it. Which is a small version of the bigger evaluation lesson: you only learn what a tool really fits for when you try to use it for something that matters.

I am going to budget for Claude Design deliberately going forward. Not just "try it." Actually plan the credits around a specific build I want to complete.

OK But What Do I Actually Do With This?

Two things. This week.

1. Test Claude Design With a Real Brief

If you have a paid Claude plan, do not waste your first session on a throwaway idea. Pick something you have been wanting to visualize. Something you have thought about enough to describe well. Feed it context. Upload a reference document. Give it a clear brief.

See what it does when you actually show up prepared.

You might still hit the credit wall like I did. But what you get before that wall will tell you whether this tool has a spot in your stack.

2. Notice the Default Aesthetic

When you do your first real build, pay attention to what the tool leans toward visually before you tell it otherwise. The fonts. The colors. The spacing. The visual energy.

Ask yourself: do I want to start from this aesthetic and push outward, or do I want to override it from the beginning?

Either answer is fine. But knowing which one fits your work helps you use the tool more intentionally.

What This Means for You Right Now

I said in Edition 347 that Claude Design might matter more for dyslexic and neurodivergent creators than anyone has noticed yet. After actually using it with a real brief, I am more convinced of that than I was before.

Not because it is the best tool on the market. Not because it solves every design problem. Not because it is finished.

Because it reduces a specific kind of cognitive load that traditional design tools create for brains like ours. And it has taste. And it collaborates cleanly with the rest of the builder stack.

For ten minutes of real use with a strong brief, I saw enough to know this belongs in my workflow.

I will spend more time with it next week when my credits reset. When I do, I will tell you what I learn. What breaks. What surprises me. What does not live up to the hype.

Because that is the deal. Honest experiments. Real results. No inflated claims. Even when the tool is as impressive as this one is turning out to be.

Previously

• Edition 347: "I Played With Claude Design for Ten Minutes" (first impressions, Anthropic Labs launch, Opus 4.7)

• Edition 346: "The Meta Layer" (builder evaluation, six-tool stack, cognitive fit)

• Edition 345: "We Have Been Asking the Wrong Question About AI" (evaluation framework manifesto)

• Edition 344: "I Woke Up at 4AM With a Random AI Idea" (Cognitive Partner OS prototypes)

• Edition 341: "I Have Never Seen Anything Like This Before" (building, state of AI)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model, HGI)

Next

Edition 349: A new academic paper just named something I have been writing about for three years. They call it the "LLM Fallacy." I have been calling it something else. The difference matters a lot for neurodivergent creators, and the Cognitive Balance Model is the answer the researchers did not know they needed.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

🎨 Ran a real experiment with Claude Design using a detailed, iterated prompt. Hit my credit wall before the build finished. What I saw before the wall was enough to tell me this tool has real potential.

✨ Claude Design has a distinct aesthetic. Soft tones. Pastel accents. Light, airy backgrounds. The kind of default taste that helps a non-designer produce work that looks considered without having to describe every visual choice.

🏗️ It fits the stack from Edition 346 as the visual and prototyping layer. Sits alongside Claude Code, Bolt.new, and Lovable. Hands off cleanly. Not a replacement. A new layer.

💰 Credits matter. Running out mid-build was a real lesson in evaluating tools. Budget deliberately for the work you actually want to do, not just for "trying it out."

🧠 For dyslexic and neurodivergent creators, the default aesthetic doing some of the heavy lifting is a real accessibility benefit. Start from tasteful, push outward. Easier than starting from blank.

🧪 Two things to do this week: test with a real brief, not a throwaway one. And notice whether the default aesthetic is a starting point you want to build from or override.

🔒 Cognitive Partner Members get the live updates as I rebuild the Cognitive Partner OS mockup with Claude Design next week. 50 founding spots at $19/month, locked forever.

Edition 347 | April 27, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• What Claude Design is and why Anthropic just launched it

• My honest, messy, minimal-prompt first impression

• Why this tool matters differently for dyslexic and neurodivergent creators

• How it fits into the builder stack from Edition 346

• What Claude Opus 4.7 means for the design quality under the hood

• Three ways to experiment with Claude Design this week

Reading Time: 8 minutes Listening Time: 12 minutes

Happy Monday.

Recently, Anthropic launched something called Claude Design out of their Anthropic Labs division. It is a new product that lets you create polished visual work through conversational prompts. Designs. Interactive prototypes. Slide decks. One-pagers. Marketing collateral. All of it through natural language.

Under the hood, it is powered by Claude Opus 4.7, which is Anthropic's newest and most capable vision model, also released the same day.

I got my hands on it last.

For about ten minutes.

I will be honest with you. I did not prompt it properly. I did not lay out a careful brief. I did not reference my existing brand or style. I just jumped in and asked for something.

And it still turned out cool.

That is the part I cannot stop thinking about. Because if it can produce something genuinely useful from me being sloppy, I can only imagine what it does when somebody actually dials it in.

What Claude Design Actually Does

Let me give you the real picture.

You describe what you need. Claude generates a first version. Then you refine through conversation, inline comments, direct edits, or custom sliders that Claude creates for your specific project.

That last one is interesting. The tool generates its own adjustment controls based on what you are building. Not a fixed interface. A custom one for your context.

It can pull from your existing design system. If your team has a codebase or design files, Claude reads them during onboarding and builds a design system automatically. Every project after that uses your colors, typography, and components. You can even have multiple design systems for different brands or products.

You can start from a text prompt, upload images or documents (DOCX, PPTX, XLSX), or point Claude at your codebase. There is a web capture tool that lets you grab elements directly from your website so prototypes look like your real product.

And then, maybe most importantly for builders like me, Claude Design work can be handed off directly to Claude Code. Which means you can prototype visually, then turn it into actual working code without the usual translation gap between designer and developer.

That closes a loop I have been watching for years.

My Ten Minutes With It

I did not have a plan.

I opened Claude Design with a cup of coffee and a rough idea that I wanted to see what it would do with almost nothing. I typed a short prompt. I did not reference my existing branding. I did not upload my Dyslexic AI style guide. I did not feed it any context about my work.

I just asked for something and watched.

What it came back with was not perfect. It was not going to replace a working product immediately.

But it was good.

Not good in the "cool demo" sense. Good in the "I could actually use a version of this" sense. The composition was clean. The spacing was intentional. The color choices were defensible. The hierarchy of information made sense without me having to explain what should be a heading and what should be body text.

That is a big deal.

Because if you have ever tried to get a traditional AI image tool to produce a usable layout, you know what a mess it usually is. Most of them either give you something that looks like clip art, or something that looks like someone had a stroke in Photoshop.

This did not do that.

And I was barely paying attention.

Which brings me to the point of this edition.

Why This Matters Differently for Dyslexic Creators

Here is the thing I want you to understand.

Most coverage of Claude Design is going to focus on how it challenges Figma and Adobe and Canva. That framing is fine. Anthropic is clearly pushing into the application layer. The business story writes itself.

But that framing misses the part that matters most to our community.

For a dyslexic or neurodivergent creator, traditional design tools are not just difficult. They are often hostile.

Figma requires you to learn a specific spatial logic. Layers. Components. Auto layout. Constraints. It is powerful if you can hold all of that in your head at once. For a brain that does not hold structured information that way by default, it can be exhausting before you have produced anything.

Adobe's suite is even harder. Each app has its own set of assumptions about how you should think. And those assumptions do not always match how a dyslexic mind organizes visual ideas.

Canva is more approachable. But it still assumes you know what you want before you start. And it rewards people who can articulate design choices in traditional design vocabulary.

Claude Design does something different.

It lets you describe what you want in the way your brain already works. Messy. Nonlinear. Voice-first if you want. You can say "make it feel calmer" and it understands what you mean. You can point at an element and say "this part is bothering me but I am not sure why" and it starts suggesting changes.

That is the cognitive fit principle I wrote about in Edition 345 applied to design.

Not a tool that forces your thinking to match its interface. A tool that adapts to how you actually think.

For dyslexic creators, that is not a convenience feature. It is an accessibility breakthrough.

The Opus 4.7 Angle Nobody Is Connecting

Here is a detail most of the coverage missed.

Claude Design is powered by Claude Opus 4.7, which Anthropic released the same day. And Opus 4.7 is specifically designed to be better at visual reasoning, dashboard creation, and design taste.

Anthropic's own announcement included quotes from people saying things like "it makes choices I would actually ship." That is not marketing speak. That is designers and builders saying a model has developed something close to aesthetic judgment.

Why does that matter for dyslexic creators?

Because a lot of us can see when something is wrong without being able to articulate why. We feel the imbalance. We sense the bad hierarchy. We know the color is off. But we do not always have the vocabulary to explain it in traditional design language.

A model with genuine design taste meets us halfway.

It can execute on a vague "this does not feel right" because it has its own sense of what "right" means. We are not trying to translate our fuzzy intuition into precise technical instructions anymore. We are collaborating with something that has intuition of its own.

This is the Cognitive Balance Model in a new context. Human Initiation: you describe a feeling or goal. AI Expansion: the tool generates multiple directions and variations. Human Integration: you react, refine, and choose.

That is a Cognitive Balance Model score of around 13 out of 15 on the HGI. Strong across all three phases, when you know how to use it.

Where This Fits in the Stack

In Edition 346 ("The Meta Layer"), I walked through the six builders I am currently evaluating for the Cognitive Partner OS project. Claude Code. Cursor. Codex. Google AI Studio. Bolt.new. Lovable.

Claude Design fits into that stack in a new way.

It is not a replacement for any of those builders. It is a visual and prototyping layer that sits alongside them. If I need a dashboard mockup, a one-pager for the Cognitive Partner membership, a prototype for the family evaluation tool, or slides for a pitch, this is now the fastest path from idea to something polished.

And because it hands off to Claude Code cleanly, it shortens the loop between "I have an idea" and "I have working software with a professional-looking interface."

For someone like me, who does not want to become a designer but wants to ship products that look like a designer was involved, this is the missing piece.

I am going to start using it this week for pieces of the Cognitive Partner OS work. I will report back on what it can do when I actually give it proper context and a real brief.

OK But What Do I Actually Do With This?

Three things.

1. Try It With a Simple Prompt First

If you have a paid Claude plan (Pro, Max, Team, or Enterprise), you have access right now. Do not overthink your first attempt. Just ask for something simple. A one-pager about something you care about. A prototype for an app idea you have been sitting on. A slide deck for a conversation you need to have.

See what it gives you when you are not trying hard.

Then ask yourself: did this remove a barrier I usually hit?

2. Feed It Your Context on the Second Try

On your second attempt, give it more. Upload a document that captures your brand voice. Point it at your website. Reference a design system you like.

Watch the output shift. That is the Single Source of Truth principle from Edition 329 applied to visual work. The more context the tool has about you, the more it produces work that actually fits you.

3. Notice What Changed in Your Body

This one sounds weird. Stay with me.

When you use a traditional design tool, notice how it feels. For a lot of dyslexic and neurodivergent users, there is a tension that shows up in your shoulders or a fog that shows up in your thinking. That is your brain working against the interface.

Now try Claude Design. Does that tension show up, or does it stay away?

If the tension stays away, you just found a tool that has real cognitive fit for you. That is worth more than any feature comparison chart.

What This Means for You Right Now

Anthropic is not just shipping features anymore. They are shipping entire new application categories.

Claude Code for developers. Cowork for knowledge workers. Claude Design for visual creators. All built on top of their frontier models. All connected to each other. All designed to work with natural language as the primary interface.

For neurodivergent thinkers who have spent decades adapting to tools that did not fit us, this is a different world.

The tools are finally starting to adapt to us.

I wrote about this in Edition 332 ("A Year Ago, I Was in a Hospital Bed") when I introduced the Cognitive Balance Model. I wrote about it again in Edition 341 ("I Have Never Seen Anything Like This Before") when I talked about the joy of building without feeling like I was fighting the interface. And now, in April 2026, I am looking at a design tool that does the same thing for visual work that Claude Code did for code and Cowork is starting to do for knowledge work.

Each of these tools lowers a barrier that used to block a specific kind of creator.

Claude Design lowers the design barrier.

For me, for Makena, for the homeschool families I have been building tools for, for the dyslexic entrepreneurs and neurodivergent professionals in this audience, that matters.

Because we do not need to become designers. We need to be able to communicate our ideas visually without the act of design becoming the thing that stops us.

Claude Design just made that easier. Ten minutes of playing with it confirmed it.

I cannot wait to see what it looks like when I actually give it a proper brief.

Previously

• Edition 346: "The Meta Layer" (evaluating builders, cognitive fit, the six-tool stack)

• Edition 345: "We Have Been Asking the Wrong Question About AI" (evaluation framework manifesto)

• Edition 344: "I Woke Up at 4AM With a Random AI Idea" (Cognitive Partner OS prototypes)

• Edition 343: "Stanford Just Measured Everything About AI" (AI Index, jagged frontier)

• Edition 341: "I Have Never Seen Anything Like This Before" (building in chaos, joy of vibe coding)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model, HGI)

• Edition 329: "Building Your Second Brain" (Single Source of Truth)

Next

Edition 348: I am going to actually give Claude Design a proper brief this week and build a real Cognitive Partner OS mockup with it. Then I will tell you what happened and share what came out. Live experiment, real results, no inflated claims.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

🎨 Anthropic launched Claude Design on April 17, powered by the new Claude Opus 4.7 vision model. Prototypes, slides, one-pagers, design work, all through conversational prompts.

⏱️ I played with it for ten minutes with almost no context. It still produced something good. Not perfect. But good in the "I could actually use this" sense.

🧠 For dyslexic and neurodivergent creators, this matters differently than the Figma/Adobe framing suggests. Traditional design tools can be hostile to how our brains work. This one adapts to us instead of forcing us to adapt to it.

🎯 Claude Opus 4.7 has something close to genuine design taste. That matters for anyone who can feel when something is off but cannot always articulate why.

🧱 It fits into the builder stack from Edition 346 as a visual and prototyping layer. Hands off cleanly to Claude Code. Shortens the loop from idea to working, professional-looking product.

🧪 Three ways to experiment: try it with a simple prompt first, feed it your context on the second try, and notice whether the usual design-tool tension shows up or stays away.

Edition 346 | April 19, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• Why building an AI evaluation tool means I have to evaluate the AI tools I build with first

• Why this recursive problem is weirder and more important than it sounds

• The six builders I am actually considering right now (and what each one does well)

• The evaluation criteria that matter when you are choosing a builder, not just using one

• Why "cognitive fit for my own brain" might be the deciding factor

• How any dyslexic thinker with an idea and a laptop can run this same decision process

Reading Time: 10 minutes Listening Time: 14 minutes

Happy Sunday.

In Edition 345 ("We Have Been Asking the Wrong Question About AI"), I left you with a cliffhanger.

I told you that if I am going to build a software product that helps businesses and families evaluate AI tools, I also need a way to evaluate the AI tools I use to build that software.

I called it meta. I promised a whole edition of its own.

This is that edition.

And I am going to tell you right up front: this is weirder and more important than it sounds.

Not because the concept is hard. It is actually pretty simple once you see it.

Because the people who understand this recursion early are going to have a huge advantage in the AI economy. And most people are still not thinking about it at all.

Let me show you what I mean.

The Problem in One Sentence

Here it is, as simply as I can put it.

Before I can build a tool that helps people choose the right AI tool for their needs, I have to choose the right AI builder for building that tool.

Yes, that is a tongue twister.

No, I did not plan it that way.

It just happens to be the honest truth about where I am sitting this morning.

Because the evaluation framework I laid out in Edition 345 requires software. Dashboards. Scoring engines. Weighted criteria. Custom templates. Export capabilities. User onboarding. Mobile responsiveness. A clean interface that does not make neurodivergent users want to throw their laptop across the room.

I cannot build that with a prompt. I need a builder.

And in April 2026, there are a lot of builders to choose from.

The Landscape Right Now

Here is what I am actually looking at. Not a theoretical list. The tools I am actively considering for the Cognitive Partner OS stack I talked about in Edition 344.

Claude Code. Anthropic's terminal-based agentic coding tool. This is my daily driver for almost everything I have been vibe coding over the last six months. It is powerful, it handles complex codebases well, and it remembers context across sessions better than most. The shipping pace right now is insane. Just in the last two weeks they have added parallel sessions, session recaps, and interactive /powerup lessons, as we covered in Edition 343.

Cursor. The AI-native IDE built on VS Code. Many developers consider it the most polished AI coding experience right now. Multi-model support. Inline completions. Agent mode. For someone doing serious engineering work, this is often the answer.

Codex. OpenAI's coding agent. I was using this yesterday when I built one of the two prototypes I mentioned in Edition 344. It got the JSON ingestion working and has solid reasoning for complex tasks.

Google AI Studio. This is where I built the other prototype yesterday. It has some interesting advantages for quick iteration and working with Gemini models directly. Not traditionally thought of as a "builder" in the same sense as the others, but for certain kinds of prototyping it is fast and accessible.

Bolt.new. I have been using Bolt since it launched, and it was my go-to tool before I got more familiar with Claude Code. It is still part of my stack. Bolt was instrumental in building most of the websites I am currently running. When I need to go from an idea to a working site fast, Bolt has been the reliable answer.

Lovable. Specifically built for rapid prototyping and stakeholder demos. If I wanted to show a family or a business what the evaluation framework looks like before building the real version, Lovable could probably get me there faster than anything else on this list.

Six builders. Six different strengths. Six different fits for different brains, different projects, different stages.

And I have to pick.

Or more accurately, I have to pick which one to start with, because most serious projects end up using more than one.

Why This Is Not Just a Technical Question

Here is where it gets interesting.

When most people write about AI coding tools, they compare them on the wrong axes.

Speed. Cost. Model support. Autocomplete quality. Enterprise features.

Those things matter. But they are not the most important thing for me.

Because I am not a traditional developer. I am a dyslexic thinker with a recreation degree who talks to his computer.

So when I evaluate a builder, the questions I ask are different.

Does it tolerate my voice-first workflow? Can I dump messy thoughts into it without being punished for imprecise syntax? Will it ask me clarifying questions when I am ambiguous, or will it just guess wrong and force me to debug its assumptions?

Does it keep context across sessions? Because if I have to re-explain the project every time I open the tool, that is cognitive load I cannot afford to pay over and over. This is the jagged frontier problem from Edition 343 applied to builders instead of end-user models.

Does it show me what it is thinking? When something breaks, can I understand why? Or does it feel like a black box that either works or does not work with no way to debug?

Does it handle error tolerance well? When I describe what I want in nonlinear, loop-heavy, voice-note language, does it still get the gist? Or do I have to clean myself up before the tool will work?

Does it let me iterate fast? Because the way my brain works, the first version is never right. The second version is closer. The fifth version is usually good. I need a builder that does not punish me for needing more cycles.

Does it fit my cognitive style? Not the abstract idea of fit. The real thing. When I use it for three hours, do I feel energized or drained?

This is the cognitive fit framework from Edition 345, applied to the tool I am choosing to build with, not just the tools I am building for others.

Every one of those questions matters more to me than speed or model size or enterprise features.

And my hunch is that they matter more to a lot of other people too. Especially neurodivergent people. Especially people who have been told their whole lives that their brains are the problem.

They are not the problem. The tools have been the problem. Now we can choose tools that fit us instead of forcing ourselves to fit the tools.

The Three-Layer Stack

Here is how I am thinking about this right now.

Layer 1. Evaluate the builders I am considering using. Claude Code. Cursor. Codex. Google AI Studio. Bolt.new. Lovable. Score them against my own cognitive fit criteria.

Layer 2. Use the winning builder (or builder combination) to create the evaluation platform itself. The Cognitive Partner OS engine for dyslexic thinkers. The business version. The family version.

Layer 3. Use that platform to help others evaluate AI tools for their own needs. Individuals. Families. Homeschoolers. Businesses. Teams. Eventually schools.

That may sound recursive. It is. But it is also practical.

Because the quality of Layer 3 depends on the quality of Layer 2. And the quality of Layer 2 depends on whether I picked the right tool at Layer 1.

Get the first layer wrong and everything above it gets harder.

This is why I keep coming back to the point that evaluation is not just a skill for end users. It is a skill for builders. For founders. For anyone who is trying to go from idea to working software in an AI-first world.

The Honest Truth About Where I Am

I want to be straight with you.

I have not made the final decision yet. I am in the middle of Layer 1.

Yesterday I got both Google AI Studio and Codex to ingest my ChatGPT JSON export. That was the first proof of concept for Engine 1 of the Cognitive Partner OS. Both of them worked. Both of them felt different. Both of them have strengths I am still figuring out.

Claude Code is probably going to be in the stack somewhere because it is already my daily driver and the new April features (parallel sessions, session recaps, team onboarding commands) are specifically designed for the kind of long-running project I am building.

Cursor is tempting for the IDE experience but I do not need a full engineering environment for most of what I do.

Bolt.new and Lovable are both interesting for the parts of this project that are user-facing. Bolt is already where most of my current websites live, so it has a built-in advantage for continuity. Lovable is the faster path for a demo-grade dashboard. If I need to ship something a family can actually use on a phone or a tablet, one of those two is probably the right answer for that layer specifically.

What I am probably going to end up with is a combination. Different tools for different layers of the same project. Claude Code for the backend logic and the evaluation engine. Bolt or Lovable for the user-facing dashboard. Google AI Studio or Codex for specific prototyping tasks where I want a different model's perspective.

The answer is not "one tool to rule them all." The answer is "the right tool for the right layer."

And I can only figure that out by running the evaluation.

This is exactly the point I made in Edition 345. The people who win in this era are not the ones chasing the hottest single tool. They are the ones who know how to evaluate fit across options and build a stack that actually works for their specific situation.

Why This Matters for Everyone (Not Just Builders)

You might be reading this thinking "Matt, I am not building software. Why should I care about this?"

Here is why.

The same recursion shows up in every important AI decision right now.

If you are picking an AI tool for your business, you are also picking the vendor ecosystem that will surround it. The integrations. The data policies. The future trajectory. The cultural fit with your team. The tool is not just the tool. It is a whole stack of choices.

If you are picking an AI tool for your homeschool, you are also picking the implicit values baked into that tool. How it handles errors. How it encourages or discourages certain kinds of thinking. How it treats a kid who does not fit the default mold.

If you are picking an AI tool for yourself, you are picking a cognitive partner that is going to shape how you think, what you focus on, and what you end up building. That is not a small decision.

Every AI tool choice is actually a stack of choices. The meta layer is real whether you are aware of it or not.

The only question is whether you are going to evaluate it deliberately, or just hope you picked right.

OK But What Do I Actually Do With This?

Three things. This week.

1. Name the Tools in Your Stack

List every AI tool you use regularly. Not just the chatbots. The ones built into your email. Your calendar. Your writing apps. Your design tools. The ones you might not think of as "AI tools" but definitely are.

You probably have more than you think. And most of them were not chosen deliberately.

2. Score Them Against Cognitive Fit

Pick the three most important ones. Score each one from one to ten on questions like: Does it fit how I actually think? Does it reduce cognitive load or add to it? Does it support voice? Does it handle my mistakes gracefully? Do I feel energized or drained after using it?

You might find out that a tool you have been using for six months is not actually the right fit. That is useful information.

3. Ask Yourself What the Next Layer Looks Like

If you are building anything (content, a business, a family system, a homeschool curriculum, a project), ask the meta question. What am I using to build the thing I am building? Is that tool the right fit? Have I evaluated it deliberately, or did I just grab the first one that worked?

That question is worth 15 minutes a week. Minimum.

What This Means for You Right Now

We are at a strange moment in AI history.

The tools are accelerating faster than anyone can track. Edition 343 covered the Stanford AI Index showing capability jumps that broke the scale. Edition 341 covered Claude's shipping pace and the sheer volume of new options launching every week. Edition 344 was me building two prototypes in one morning.

The pace is not slowing down. It is speeding up.

Which means the people who can evaluate tools well, at every layer of their stack, are going to compound advantage every month. And the people who just grab whatever is trending are going to find themselves repeatedly migrating, repeatedly frustrated, repeatedly starting over.

Sticktoitness is not just about discipline. It is about conditions, as Tobin Trevarthen said in the piece I linked in Edition 342. The tools you choose are part of your conditions. Choose badly and you are fighting your environment. Choose well and the environment supports you.

I am choosing right now. In the open. With you.

And I will tell you what I land on, why, and what I learn in the process. That is the deal.

Previously

• Edition 345: "We Have Been Asking the Wrong Question About AI" (evaluation framework manifesto, one engine two tracks, the cliffhanger)

• Edition 344: "I Woke Up at 4AM With a Random AI Idea" (Cognitive Partner OS, the first prototypes)

• Edition 343: "Stanford Just Measured Everything About AI. They Forgot to Measure Us." (AI Index, jagged frontier, Stanford dyslexia research)

• Edition 342: "The Weight in My Chest" (sticktoitness, conditions, autonomy)

• Edition 341: "I Have Never Seen Anything Like This Before" (state of AI, building evaluation tools)

• Edition 340: "I Have Four of the Five Layers. Time to Close the Loop." (self-improving loop)

• Edition 339: "Your AI Just Forgot Everything. Again." (Karpathy, five-layer stack)

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

🔁 Before I can build a tool that helps others evaluate AI, I have to evaluate the AI tools I use to build that tool. Meta, but real.

🛠️ Six builders in the running: Claude Code, Cursor, Codex, Google AI Studio, Replit, Lovable. Each strong at different things. None of them wins on every axis.

🧠 For dyslexic thinkers, the evaluation criteria are different. Not speed or model size. Cognitive fit. Voice tolerance. Error handling. Iteration speed. Whether three hours with the tool leaves you energized or drained.

🏗️ The three-layer stack: evaluate the builders, pick the best fit, use the winning tool to build the platform that helps others evaluate AI tools for their own needs. Get Layer 1 wrong and everything above it gets harder.

🤝 The answer is usually not one tool. It is a combination. Different tools for different layers of the same project.

🧩 Every AI tool choice is actually a stack of choices. The meta layer is real whether you are aware of it or not. The only question is whether you evaluate deliberately or hope you picked right.

🔧 Three things to do this week: name all the AI tools in your stack, score the top three on cognitive fit, and ask what the next layer looks like for whatever you are building.

Edition 345 | April 18, 2026The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• Why most people are asking the wrong question about AI tools

• Why dyslexic and neurodivergent thinkers need a completely different benchmark

• What a real dyslexic AI evaluation framework should measure

• Why businesses should build their own internal AI scorecard instead of copying everyone else

• Why families, homeschoolers, and non-traditional learners need their own version too

• How one evaluation engine could serve three very different users

• Why evaluation itself might be the most important skill of the AI era

Reading Time: 11 minutes
Listening Time: 15 minutes

Happy Sunday.

This morning I found myself going down a rabbit hole that felt very familiar.

It started with a simple question.

Which AI tool is best?

That sounds like a smart question. It sounds practical. It sounds like what everyone wants to know right now.

But the more I sat with it, the more I realized it is still the wrong question.

Because best for who?

Best for what?

Best in what situation?

Best for a dyslexic thinker? Best for a parent trying to homeschool? Best for a small business owner trying to save time? Best for a team that needs structure and compliance? Best for a kid who learns better by talking than typing? Best for a founder whose brain works in loops, patterns, and voice notes instead of neat little spreadsheets?

That is where this whole thing started opening up.

And once I followed that thread, I realized we may be building not just one idea, but an entire framework.

The Real Problem Is Not the Tool

The real problem is that most people are trying to choose AI tools before they even know how they should evaluate them.

That is backward.

Right now, most people are buying AI the same way people used to buy supplements, software, or shiny new productivity hacks. They hear the hype. They see a cool demo. They try the newest thing. They compare headlines. They ask who has the smartest model.

But that still skips the most important part.

What does "good AI" actually mean for your life, your work, your brain, your child, your business, or your family?

Because raw intelligence alone is not enough.

A model can be powerful and still be a bad fit. It can be too text-heavy. Too cluttered. Too rigid. Too hard to steer. Too weak on voice. Too generic. Too overwhelming. Too complicated for real life.

So the deeper question is not "which AI is smartest?"

The better question is:

Which AI is the best fit for how you think, learn, work, and live?

That is a very different category. And I think it matters a lot.

Dyslexic Thinkers Need a Different Benchmark

This is the part that matters most to me personally.

As a dyslexic thinker, I know firsthand that the smartest tool is not always the most usable tool.

That may sound obvious, but it is not how most people talk about AI.

Most AI rankings focus on reasoning, coding, speed, benchmark scores, model size, and enterprise features. Those things matter. In Edition 343 ("Stanford Just Measured Everything"), we looked at the AI Index and saw exactly what the industry measures: capability on academic tests, adoption rates, patent filings, investment dollars.

But from a dyslexic AI perspective, I would add a completely different set of questions.

Does the tool reduce cognitive load? Does it let me speak instead of type? Does it help me organize messy thinking? Does it make complex information easier to understand? Does it chunk ideas clearly? Does it work with nonlinear thought instead of fighting it? Does it help me feel more capable, not less? Does it support confidence, clarity, and action?

That is a different kind of benchmark.

This is not a benchmark for intelligence. This is a benchmark for cognitive fit.

That phrase matters. Because I believe the future of AI is not just about how intelligent the tool is. It is about how usable that intelligence is for different types of minds.

A tool can be brilliant and still be exhausting.

A tool that reduces overwhelm, supports voice, tolerates messy prompts, and helps people stay in motion may actually be the better tool for many dyslexic and neurodivergent users, even if it does not win every mainstream benchmark.

That is a huge shift. And it connects directly to what Stanford called the jagged frontier in Edition 343. The idea that AI is brilliant at some tasks and bafflingly bad at others. If that is true of the tools, it is doubly true of the fit between a tool and a specific brain.

What a Dyslexic AI Benchmark Should Actually Measure

If I were building a benchmark specifically for dyslexic thinkers, and as I mentioned in Edition 341, I have already started, I would not just rank models by "best answer." I would score them based on how well they support real human thinking.

Some of the categories I would want to measure:

1. Voice-First Support. Can I talk to it naturally? Can it understand me well? Can it help me think out loud? Can it read things back to me clearly?

2. Reading Load. Does it give me a giant wall of text? Or does it break ideas into usable chunks?

3. Writing Support. Can it help me take messy thought and turn it into clear language without flattening my voice?

4. Error Tolerance. Does it handle typos, fragmented prompts, and nonlinear input well?

5. Clarity. Can it explain something in normal human language instead of jargon soup?

6. Adaptability. Can it adjust to the way I like to learn, process, and communicate?

7. Cognitive Load Reduction. Does it save me mental energy, or does it create more work?

8. Confidence Support. After using it, do I feel more capable and more willing to keep going?

This is where I think Dyslexic AI has a real opportunity.

Not just to talk about AI. But to help define how AI should be evaluated for people whose brains do not work in the default factory setting the world tends to optimize for.

If you have been following since Edition 332 ("A Year Ago, I Was in a Hospital Bed"), you know the Cognitive Balance Model was built to respect how neurodivergent minds actually work with AI. The Human Guidance Index measures collaboration quality. These frameworks already exist. What is missing is a standardized evaluation lens applied to every tool on the market through this filter.

That is the opportunity.

Then the Idea Got Bigger

Once I started thinking about benchmarks for dyslexic thinkers, I realized this same logic applies to businesses too.

Because businesses are asking the same wrong question.

They keep asking which model is best. Which AI platform they should use. Which tool they should buy. Which agent they should install.

But they are still skipping the first step.

Before a business chooses AI tools, it should build its own internal evaluation system.

It should ask: What are our actual use cases? Where are we wasting time? What are our biggest bottlenecks? What kind of outputs do we need? What matters most to our workflow? What risks do we care about? What kind of team are we supporting? What does success actually look like here?

Every business should have its own AI evaluation framework. Not a generic list from the internet. Not a copy of what another company is doing. Not a hype-driven guess.

A real scorecard based on its own needs.

The Business Version

This is where LM Lab AI comes in.

The business version of this concept would help a company create its own internal benchmark for AI adoption. In simple terms, it would help a business define what "good AI" means for them before they waste time or money choosing tools.

That system could guide them through questions like: What department are we evaluating for? Sales, operations, customer service, HR, education, leadership? Are we trying to save time, improve quality, reduce overload, or increase consistency? Do we need speed, accuracy, explainability, privacy, integrations, or ease of use? What tasks should we test? How should we compare tools? What weights should we give each category?

Then the software or consulting process could generate custom evaluation criteria, weighted scorecards, model comparison templates, test scenarios based on real workflows, recommendation reports, and pilot rollout suggestions.

That turns AI adoption into something much more thoughtful and useful.

Not hype. Not chaos. A framework.

In Edition 334 ("The Data Is In"), we looked at the Anthropic labor market study that showed cognitive flexibility as the new job security. Businesses with evaluation frameworks are more cognitively flexible by design. They are not locked into one tool or one vendor. They have a repeatable process for assessing new options as the landscape changes. And it changes a lot. Edition 341 covered the insane pace of releases just in the last two weeks.

Most businesses do not need more AI noise. They need structure, literacy, priorities, and a way to make better decisions.

This Is Not Just a Business Tool

Then I realized something else.

This same process could work for families too.

And maybe that is the part that hits closest to home.

Because in my daily life, working through tools for my own kids, my own family, and my own educational experiments, I know how confusing this space can be.

Families have more freedom than schools in a lot of cases. But they also have less guidance. They can choose almost any AI tool they want, but they still do not know which ones are actually helpful, which ones are good for learning, which ones are good for reading support, which ones are safe enough, which ones work well for voice, which ones can support executive function, which ones are age appropriate, or which ones help without creating more screen addiction or mental clutter.

That is a real problem.

Especially for homeschool families. Neurodivergent learners. Dyslexic kids. Parents trying to use AI intentionally. Non-traditional education settings. Families that want more agency and customization.

So now this idea is not just a business tool. It is also a family and learning fit tool.

If you have been here since Edition 325 ("My 14-Year-Old Daughter Just Proved Me Wrong"), you know this is personal. My daughter Makena is homeschooled. I build her tools. I make these evaluation decisions every week. And every week I see how much harder it is than it needs to be.

Families Need Their Own Version of "Best"

A family should not evaluate AI the same way a corporate operations team does.

That sounds obvious. It matters more than most people realize.

A family or homeschool version of this framework would ask different questions.

Does this tool help my child understand concepts? Does it support reading differences? Does it help with writing without taking over? Does it work well with voice? Is it easy for a parent to manage? Does it support curiosity and confidence? Can it adapt to different ages and subjects? Does it help with executive functioning, routines, and planning? Does it fit our values, our daily rhythm, and our educational style?

Very different evaluation system. Same core engine. Different lens.

That may be one of the most useful applications of all of this. Schools are often limited by policies, budgets, restrictions, and slow decision-making. Families have more autonomy. That means they can become one of the earliest proving grounds for better AI fit.

The Stanford dyslexia research we covered in Edition 343 showed that the right tools, applied the right way, physically rewire the dyslexic brain. That finding cannot stay in a lab. It needs to reach homes. And it needs a framework for choosing those tools intentionally.

One Engine, Two Tracks

At one point this morning, I had to stop and ask myself a real question.

Are these two separate products? Or is this one product with two lines of business?

The answer I came to is this.

It is one core product with two front-end tracks.

Pathway 1: Family, Personal, Homeschool.
Pathway 2: Business, Team, Organization.

Under the hood, both would be doing the same thing. Understanding goals. Identifying pain points. Building evaluation criteria. Assigning weights. Comparing tools. Generating recommendations. Creating an action plan.

That is the core engine. What changes is the language, onboarding, templates, scoring priorities, and outputs.

That feels like a very strong product direction. Not two totally different companies. Not two giant separate software projects. One evaluation engine. Two applied markets.

This also ties directly to what I was building in Edition 344 ("I Woke Up at 4AM With a Random AI Idea"). The Cognitive Partner OS. The underlying engine for understanding how you actually use AI. That tool and this evaluation framework are siblings, not rivals. One maps your current usage. The other helps you choose tools that fit.

Both are powered by the same core insight. Fit matters. And you cannot measure fit without knowing yourself first.

The Meta Layer (A Cliffhanger)

Just when I thought we had the framework, I realized there was one more level to it.

If I am going to build a software product that helps businesses and families evaluate AI tools, then I also need a way to evaluate the AI tools I use to build that software.

Yes, I know. That is very meta.

It is also true. And it is a whole edition of its own.

I will get into that recursive layer another time. For now, just know that the same principles apply. Speed to prototype. Logic flexibility. Cognitive fit for my own brain. The tools we use to build the evaluator need to pass the evaluator's own test.

Save that thought. We will come back to it.

The Bigger Point

This is what I am circling around.

The future is not going to belong only to the people who know how to use AI.

It is going to belong to the people who know how to evaluate it well.

That means knowing what matters. Knowing what to test. Knowing what outcomes you actually care about. Knowing how your brain works. Knowing how your team works. Knowing how your child learns. Knowing what "fit" means in context.

That is a much more useful skill than chasing the newest tool every week.

And I think this is where Dyslexic AI and LM Lab AI both have a lane. Not just teaching people how to prompt. Not just reviewing tools. But helping define the standards by which tools should be judged in the first place.

That is a more important game.

OK But What Do I Actually Do With This?

Three things. This week.

1. Define Your "Best" Before You Shop

Before you try another AI tool, write down five to ten things that "good AI for me" actually means. Voice-first? Clear outputs? Low cognitive load? Specific to your work? Good memory? Pick the ones that matter and write them down. This is the seed of your personal evaluation framework.

If you already have a Single Source of Truth from Edition 329, add a section called "What Good AI Looks Like for Me." Three bullet points. Done.

2. Score One Tool You Already Use

Pick an AI tool you use regularly. Score it from one to ten on each of your criteria. Be honest.

You might be surprised. You might find out your favorite tool is actually not serving you well on the things you said matter most. That is useful data.

3. Apply the Lens Somewhere It Matters

If you run a business, ask your team what criteria they would use to evaluate AI tools for your specific workflow. Not what vendors are selling. What you actually need.

If you are a parent, do the same for your family or homeschool. What does good AI for your kid look like? Now you have a framework, not a hunch.

What This Means for You Right Now

AI is moving so fast that people are understandably overwhelmed.

Too many tools. Too many claims. Too many demos. Too many experts telling everyone to jump in.

But very few people are stopping to ask the foundational question.

What should we actually be measuring?

That question matters for dyslexic thinkers. It matters for families. It matters for homeschool environments. It matters for schools. It matters for businesses. And it even matters for those of us trying to build the next generation of tools ourselves.

For me, this is one of those moments where a bunch of threads I have been circling for a long time suddenly started snapping into focus.

Dyslexic AI is not just about using AI. It is about helping define what better AI fit looks like for different kinds of minds.

LM Lab AI is not just about building tools. It is about helping people and organizations make smarter, more human-centered decisions about the tools they choose.

And the deeper I get into this work, the more I believe this may become one of the most important conversations in the next phase of AI.

Not which tool is smartest.

But which tool actually fits.

And how do we know?

That is the question I want to keep building around.

Next

Edition 346: The meta layer. How to evaluate the AI tools you use to build the tools that evaluate AI tools. Yes, really. This is the recursive problem underneath everything I am working on right now, and it is weirder and more important than it sounds.

Matt "Coach" Ivey
Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

❓ Most people are asking the wrong AI question. "Which tool is best?" skips the real question: best for whom, for what, and for how you actually think and work.

🧠 Dyslexic thinkers need a different benchmark. Not raw intelligence. Cognitive fit. Voice support. Low reading load. Error tolerance. Clarity. Confidence support.

🏢 Businesses are making the same mistake. They should build their own internal evaluation framework before choosing tools. Not copy someone else's scorecard.

👨‍👩‍👧 Families, homeschoolers, and non-traditional learners need their own version too. Same engine. Different lens. Different priorities.

⚙️ One evaluation engine. Two tracks. Personal/family on one side. Business/team on the other. Same core. Different applications.

🔮 The future belongs to the people who know how to evaluate AI well, not just the people who use it. This might be the most important skill of the next phase.

🧩 Three things to do this week: define what "good AI" means for you in five bullet points, score one tool you already use against those criteria, and apply the lens somewhere it matters.

Edition 344 | April 17, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• The random 4AM thought that kicked off this whole project

• Why I want a tool that analyzes my own AI history (and why you might too)

• The three engines I think this platform could eventually include

• What I actually got working this morning (and what I did not)

• Why this matters for dyslexic, neurodivergent, and lateral thinkers

• Three things you can do with your own AI history this week

Reading Time: 8 minutes Listening Time: 12 minutes

Sometimes my best newsletter idea is just telling you what I actually worked on.

Today is one of those days.

I did not wake up with a polished topic. I woke up at 4AM with one of those half-awake, half-dreaming thoughts that hits you out of nowhere and will not leave you alone.

The thought was basically this.

What if I could build a tool that takes all of my ChatGPT history, all of my conversations, all of my back-and-forth thinking, and actually helps me understand how I use AI?

Not just search it. Not just scroll through old threads.

Really analyze it.

What topics do I come back to over and over?

How often am I using AI to brainstorm versus write versus research versus build? What words do I repeat most? What concepts keep resurfacing? What unfinished ideas am I clearly obsessed with?

And maybe even more importantly: what does all of that say about how I think?

That one random thought turned into a couple hours of work this morning. And by the end of it, I had already built the first version of the idea in two different tools.

One in Google AI Studio. One in Codex. I have not tried it in Claude Code yet because my MacBook Air was already running hot with too many things open at once. That will probably be version three.

But I did get both versions to upload the JSON file from my ChatGPT history.

That may sound small. It was a huge first step. The concept is real enough to start.

The Original Question

At first, this started as a simple question.

If these AI platforms are already learning from how we interact with them, and all these companies are already tracking clicks, searches, prompts, time spent, behavior, and patterns, why should we not be able to use that same kind of logic for ourselves?

Why should all of that behavioral value belong only to platforms, advertisers, algorithms, and recommendation engines?

Why can I get custom ads, custom feeds, custom recommendations, and custom content ranking from companies trying to influence me, but not have a custom system that helps me understand myself better?

That is what really got me going.

In Edition 342 ("The Weight in My Chest"), I talked about wanting autonomy, agency, and sovereignty over how AI shapes my life. This is that. Not in the abstract. In code. This morning.

Because I do not want infinite customization. I do not think most people do. Most people do not want to build their own software from scratch any more than they want to code their own website.

They want something that learns them. Adapts to them. Helps them without drowning them in settings.

That is the difference. And that led me to a bigger product idea.

The Bigger Vision

The working name for this, and I want to stress that it is a working name that could change, is Cognitive Partner OS. It is related to but distinct from the Cognitive Partner Membership I launched in Edition 333. Think of it as the underlying engine rather than the community.

The idea is one platform with multiple engines inside it. Not a million random features. Not a giant mess. A real system with modular parts.

Engine 1: Conversation Intelligence

This is the first engine and the one I started with today.

Its job is to take exported AI history, ingest the JSON file, normalize the data, and analyze it through what I am calling a cognitive partner lens. It would help me answer questions like:

What projects have I returned to the most? What words and concepts do I repeat? How often do I use AI for writing, strategy, research, or technical building? What ideas keep resurfacing? What is exploratory thinking versus explanatory writing? How much of my interaction is me versus the tool?

That last one is especially important.

I want to separate my inputs from the tool's outputs. My words. My prompts. My ideas. My phrasing. Versus what the model gives back.

That becomes incredibly valuable if I want to build books, white papers, newsletters, podcasts, and all the rest from my own thinking. This is the natural extension of the Single Source of Truth framework from Edition 329. Not just a document that tells your AI who you are. A system that shows you who you are based on everything you have already said to your AI.

Engine 2: Publishing Intelligence

If Engine 1 helps me understand my thinking, Engine 2 helps me publish it.

This would be a publishing team of agents working together across books, white papers, newsletters, podcasts, audiobooks, and eventually video.

The publishing engine could pull recurring ideas, my strongest phrases, quote candidates, research themes, unfinished concepts, and connections between the book and newsletter.

Instead of starting from scratch every time, it would help me build from my own existing thinking and content ecosystem. Three years. 344 editions. That is a lot of thinking to mine.

Engine 3: Workflow Calibration

This is a longer-term idea, but it may be the most interesting.

What if the system could observe how I actually work? What tabs I open. What tools I use. How I move through a workflow. How I research, write, rewrite, post, and publish.

Not in a creepy way. Not spying. Not stealth.

A tool I own and turn on and off. An observe mode. A learning layer. A personal assistant that watches how I do my newsletter or content workflow and helps me recreate parts of that later.

Not just understanding my conversations with AI. Understanding my workflow with AI.

That is where this gets really powerful. And it ties directly back to Edition 340, where we built the self-improving loop. This is the same idea at a system level.

What I Actually Got Working This Morning

I need to be honest here.

This is still early. I do not have a magical finished product.

What I got done today was the beginning.

I built two versions of the idea. I got both versions to the point where they could finally upload the JSON file from my ChatGPT history. That took a few iterations. A little debugging. A few moments of "why is this not working?" And then finally, both versions could see the file and begin the process.

That matters. Because this is how these things start.

The first win is not perfection. The first win is proving the concept has a pulse. And today, I got it to breathe.

It was not yet doing enough with the data to give me all the insights I want. It was not yet surfacing enough specifics or value. But it could ingest the file and start the process.

That is the bridge from idea to product.

And all of it started from a random thought between sleep and wakefulness at 4AM. That still kind of blows my mind.

Why This Matters for How I Think

This project connects a lot of the things I have been thinking about for a long time.

Voice-first tools. Cognitive partner AI. Adaptive systems. Data sovereignty. Neurodivergent accessibility. Publishing. Personal dashboards. Agents. Custom software that adapts to the human instead of forcing the human to adapt to the software.

That is the larger story here.

I do not just want better prompts. I do not just want another chatbot. I want a system that helps me understand how I think, how I work, what I keep coming back to, and where I should go next.

And I think a lot of other people would want that too.

Especially dyslexic thinkers. Especially neurodivergent thinkers. Especially people whose minds are full of patterns, ideas, loops, unfinished projects, and a thousand tabs open at once.

This kind of system could help turn that chaos into structure without killing the creativity.

That is the sweet spot.

In Edition 343, we talked about Stanford's "jagged frontier" finding. The idea that AI capability is spiky, not smooth. Brilliant at some things. Bad at others. And I pointed out that dyslexic minds have been living on a jagged frontier our whole lives.

A tool like this could help us map our own jagged frontier. Where are our peaks? Where are our valleys? What do we keep coming back to that we have not finished? What patterns in our thinking are invisible to us but obvious once the data is organized?

That is not a feature. That is a mirror.

The Part I Keep Coming Back To

The more I think about it, the more I believe the real value is not just importing data.

It is taking back the algorithm for ourselves.

That might be the simplest way to say it.

Big platforms already analyze how we use their tools. They already use our behavior to shape feeds, ads, recommendations, and content.

What I want is the reverse.

I want to use my own data to shape my own cognitive environment. My own workflow. My own learning. My own publishing. My own dashboard. My own AI partner.

That starts to feel like a real future.

Not just more software. Better software. More human software. More personalized software. More adaptive software. More useful software for people who think differently.

In Edition 342, I said I refuse to wait for someone else to tell me how AI changes my life. This is what that looks like in practice. It looks like waking up at 4AM and building the first version of a tool by noon.

OK But What Do I Actually Do With This?

You do not need to build software. But you can start the same kind of self-analysis today with three simple steps.

1. Export Your AI History

Most AI platforms let you export your data. In ChatGPT, go to Settings and then Data Controls. In Claude, check your account settings for data export. The JSON file is usually easy to download and contains a lot of what you have actually said to your AI over time.

You do not have to do anything with it yet. Just have it. It is your data. Keep a copy.

2. Ask Your AI to Analyze a Small Sample

Pick a recent conversation. Paste it into a new chat and ask this:

You will be surprised what shows up. This is a mini version of Engine 1, no code required. And if you have been using your Single Source of Truth from Edition 329, you have a head start.

3. Notice What You Keep Returning To

For one week, at the end of each day, write down what topics you opened AI for. Just a list. No judgment.

By the end of the week, you will see patterns. The real ones. The ones you are too close to notice in the moment.

That is the beginning of your own Cognitive Partner OS. No engineers required.

Question for You

If you could upload your full AI history into a tool and have it show you how you think, what would you want to know first?

Your most repeated ideas? Your biggest unfinished projects? Your strongest phrases? Your blind spots? Your best insights?

Hit reply and tell me. I genuinely want to hear. Because this kind of tool is going to matter a lot more in the future than most people realize. And the people who help shape what gets built are the ones who describe what they actually need.

What This Means for You Right Now

I did not have a newsletter topic this morning.

Then I had a random 4AM thought.

A couple hours later, I had two prototype versions of a tool I have wanted for a long time without fully realizing it.

That is the kind of thing AI makes possible now. You can go from a half-formed idea to a working first version in a morning. Not finished. Not polished. Not ready for the world. But real enough to test. Real enough to improve. Real enough to believe in.

That is where I am today. And honestly, I love days like this.

Because this is what building in public looks like for me. Messy. Fast. A little obsessive. A little chaotic. Deeply exciting. And very, very real.

If this makes sense to you, if the idea of taking back your own algorithm lights something up, you are exactly the kind of person this is being built for.

Previously

• Edition 343: "Stanford Just Measured Everything About AI. They Forgot to Measure Us." (AI Index 2026, jagged frontier, Stanford dyslexia research)

• Edition 342: "The Weight in My Chest" (being understood, autonomy, sticktoitness, sovereignty)

• Edition 341: "I Have Never Seen Anything Like This Before" (state of AI, ArtQuest, building in chaos)

• Edition 340: "I Have Four of the Five Layers. Time to Close the Loop." (self-improving AI loop)

• Edition 339: "Your AI Just Forgot Everything. Again." (Karpathy, five-layer stack, memory architecture)

• Edition 333: "25 Tools. Zero Memory." (Cognitive Partner Membership launch)

• Edition 329: "Building Your Second Brain" (Single Source of Truth)

Next

Edition 345: We have been asking the wrong question about AI. "Which tool is best?" skips the real question entirely. This is the manifesto for why evaluation is the most important skill of the AI era, and why dyslexic thinkers, businesses, and families all need their own framework. Plus a cliffhanger about the meta layer underneath everything.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

⏰ Woke up at 4AM with a random thought: what if I could analyze all my AI history to understand how I actually think? By noon I had two working prototypes.

🧠 The working name is Cognitive Partner OS. Three engines: Conversation Intelligence (understand how I use AI), Publishing Intelligence (turn my thinking into books, newsletters, podcasts), and Workflow Calibration (learn how I actually work).

🔒 The core idea: take back the algorithm. Big platforms already analyze how we use their tools. This is the reverse. Use your own data to shape your own cognitive environment.

🛠️ Built in Google AI Studio and Codex this morning. Claude Code version coming when my Mac cools down. Both versions ingest the JSON file. Not finished. Real enough to start.

🧩 For dyslexic and neurodivergent thinkers: this is a tool for mapping your own jagged frontier. Where are your peaks? Your valleys? What do you keep returning to that you have not finished?

📨 Three things you can do today: export your AI history, ask your AI to analyze a small sample for patterns, and track what you keep coming back to for a week.

Edition 343 | April 16, 2026The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• What the Stanford HAI 2026 AI Index actually says (and what it misses)

• Why a model that aces PhD-level science but cannot read a clock is the most dyslexic thing in AI

• The stat that proves you are not early anymore

• What Claude shipped this month and why the pace matters

• How the same university studying AI also proved that dyslexic brains can be rewired by the right tools

• Why trust is collapsing and what that means for people who build their own systems

Reading Time: 9 minutes
Listening Time: 13 minutes

Every year, Stanford University drops a report that is basically the physical exam for the entire AI industry.

The 2026 AI Index landed this weekend. 423 pages. Nine chapters. Produced by the Institute for Human-Centered Artificial Intelligence, not by any AI company with something to sell. It is the closest thing we have to an honest, independent snapshot of where AI actually stands.

I read it. And I have thoughts.

Because Stanford measured everything. Model performance. Investment dollars. Adoption rates. Patent filings. Public trust. Job displacement. Global competition.

But they did not measure what AI means for the people whose brains work differently.

Not a single chapter on neurodivergent cognition. Not a single data point on how these tools affect dyslexic thinkers, ADHD minds, or any of the roughly 15 to 20 percent of the population that processes information outside the neurotypical default.

Which is ironic. Because buried inside this report is one of the most dyslexic findings in the history of AI research.

The Jagged Frontier (Or: My Brain Finally Has a Metaphor)

Here is the headline that made me put my coffee down.

The same AI models that now outperform humans on PhD-level science questions, competition-level mathematics, and multimodal reasoning can only read an analog clock correctly 50.1 percent of the time.

Stanford calls this the "jagged frontier." The idea that AI capability is not a smooth line going up. It is spiky. Brilliant in some areas. Bafflingly bad in others. The format of the question matters as much as the difficulty.

If you are dyslexic, you just read that and thought: welcome to my life.

We have been living on a jagged frontier since childhood. Brilliant at pattern recognition, spatial reasoning, big-picture thinking, creative problem-solving. Struggling with spelling, sequential processing, reading out loud in class.

Not because we lack intelligence. Because the format of the test does not match the shape of our thinking.

In Edition 332 ("A Year Ago, I Was in a Hospital Bed"), I introduced the Cognitive Balance Model because I needed a framework that respected how neurodivergent minds actually work with AI. Not a smooth, linear process. A spiky one. Strong in some phases. Needing support in others.

Stanford just gave that concept a name. The jagged frontier. And they applied it to the most advanced AI systems on the planet.

The machines have the same problem we do. The difference is that nobody calls the machines broken.

You Are Not Early Anymore

Here is a number that should change how you think about where you stand.

53 percent of the global population has used generative AI within three years of its mainstream launch. That is faster adoption than the personal computer. Faster than the internet.

88 percent of organizations now use AI in some form.

4 out of 5 university students use generative AI.

You are not early anymore. If you are reading this newsletter, you are not on the bleeding edge. You are in the middle of the biggest technology adoption in human history.

The question is no longer "should I use AI?" The question is "am I using it well enough?"

In Edition 334 ("The Data Is In"), we looked at the Anthropic labor market study and the conclusion was clear: cognitive flexibility is the new job security. The Stanford report backs that up. The people who adapt, who learn the tools, who build systems around how they think, those are the people who will thrive.

And in Edition 342 ("The Weight in My Chest"), I said something that I meant: I refuse to wait for someone else to tell me how AI changes my life. I want autonomy and agency over those choices.

The Stanford data says that instinct is right. Because the gap between the people who are shaping AI and the people who are being shaped by it is getting wider every month.

The Trust Problem (And Why It Is Our Opportunity)

Here is the finding that should concern everyone.

73 percent of AI experts view AI's impact on jobs positively. But only 23 percent of the general public agrees.

The US ranks dead last among surveyed nations in public trust in its own government to regulate AI. Just 31 percent.

Stanford's transparency scores for AI companies have collapsed from 58 to 40 over the past year. The biggest companies are disclosing less about their models, not more.

So the tools are getting better. But the trust is getting worse.

For most people, that is a reason to hesitate. To wait. To let someone else figure it out.

For us? It is a reason to build our own systems.

This is exactly why I have been pushing the Single Source of Truth since Edition 329. Why I introduced the Human Guidance Index in Edition 332. Why I started building evaluation tools in Edition 341.

Because if you cannot trust institutions to regulate AI for you, and you cannot trust AI companies to be transparent about how their models work, the only option left is to build your own infrastructure for understanding and controlling how you use these tools.

That is not paranoia. That is agency. And it is exactly what the Cognitive Balance Model is designed for. Human Initiation. AI Expansion. Human Integration. You stay in control at every phase.

What Claude Shipped This Month (And Why It Matters)

While Stanford was measuring the industry, Anthropic was shipping.

Just in April, Claude has released a redesigned desktop app with parallel session management. A new sidebar for organizing multiple active sessions. A session recap feature that gives you context when you come back to a conversation. Team onboarding commands. Interactive lessons called /powerup that teach you features with animated demos. Prompt caching controls. Memory improvements.

And that is just this month.

If you have been following along since Edition 333 ("25 Tools. Zero Memory."), you know that memory has been the bottleneck. The tools were smart enough. They just could not remember who you were.

That is changing. Fast. Claude's memory is now available to all users. The session recap feature addresses the exact problem we talked about in Edition 339: coming back to a conversation and having to rebuild context from scratch.

These are not flashy announcements. They are infrastructure. The boring, essential plumbing that makes AI actually useful for daily work instead of a novelty you have to re-train every session.

For neurodivergent users, this kind of infrastructure matters more than it does for anyone else. Because every time you have to re-explain yourself to a tool, that is cognitive load. And cognitive load is the tax that dyslexic thinkers pay every day just to operate in systems that were not designed for us.

Less re-explaining means less tax. More building. More creating. More of the fun I talked about in Edition 341.

The Stanford Connection Nobody Is Making

Here is the part that made me want to write this edition immediately.

The same Stanford University that produces the AI Index also runs the Stanford Reading and Dyslexia Research Program. Led by Dr. Jason Yeatman. Spanning the School of Medicine and the Graduate School of Education.

In February 2026, just two months before the AI Index dropped, Yeatman's team published a study in Nature Communications that showed evidence-based reading intervention physically changes the dyslexic brain. A six-year randomized controlled trial found that children who received the right kind of structured instruction actually grew the brain region responsible for word recognition.

Children who did not receive intervention showed no comparable change.

Let me say that again. The right tools, applied the right way, literally rewire the dyslexic brain.

Now hold that thought and put it next to the AI Index.

Stanford is telling us two things at the same time. AI capability is accelerating faster than any technology in history. And the dyslexic brain can be physically transformed by the right kind of structured support.

Nobody is connecting these two findings.

Nobody at Stanford is asking: what happens when you give a dyslexic thinker whose brain is wired for pattern recognition, spatial reasoning, and creative problem-solving access to AI tools that handle the sequential, text-heavy, memory-intensive tasks that have always been their bottleneck?

What happens when you pair a rewirable brain with tools that are improving every month?

I think the answer is what I have been writing about for 343 editions. Cognitive partnership. Not AI replacing human thinking. AI complementing the specific shape of neurodivergent thinking. Filling in the valleys of the jagged frontier while the peaks do what they have always done.

That is the research I want to see. That is the chapter missing from the AI Index. And that is why the work we are doing here matters, even when it feels like nobody is watching.

OK But What Do I Actually Do With This?

Three things.

1. Read the Jagged Frontier as a Mirror.

The next time you feel bad about what you cannot do, remember that the most advanced AI models on earth ace PhD-level physics and cannot read a clock. Capability is not a straight line. Not for AI. Not for you. Your jagged frontier is not a flaw. It is a feature.

2. Check Your AI Infrastructure.

Stanford's data says 88 percent of organizations are using AI. But how many are using it well? Open your Single Source of Truth (or build one if you have not yet, see Edition 329). Run the weekly debrief from Edition 340. Make sure your tools are working for you, not just running in the background.

If you use Claude, check out the new session recap feature. It is designed for exactly the problem we have been talking about: coming back to a conversation without losing context.

3. Demand Better Research.

If you work in education, research, or neurodivergent advocacy, push for the intersection of AI and neurodivergent cognition to be studied. Stanford has both programs under the same roof. The AI Index and the Reading and Dyslexia Research Program exist at the same university. Someone needs to connect them. Maybe that someone is in this room.

What This Means for You Right Now

Stanford measured everything about AI except what matters most to us.

That is not a criticism. It is an invitation.

The data says capability is accelerating. Trust is collapsing. Adoption is mainstream. The jagged frontier means that no model is good at everything, and the format of the task matters as much as the difficulty.

Every single one of those findings validates what this newsletter has been building toward for three and a half years.

The Cognitive Balance Model is how you navigate the jagged frontier. The Single Source of Truth is how you build your own trust infrastructure when institutions will not do it for you. The Human Guidance Index is how you measure whether you are actually in control.

And the Stanford dyslexia research? That is the biological proof that the right tools, applied the right way, change the brain itself.

We are not waiting for Stanford to make the connection. We are making it ourselves.

Nobody is getting replaced by AI. People will get replaced by other people who know how to use AI.

And the people whose brains were built for the jagged frontier? We have been training for this our whole lives.

Previously

• Edition 342: "The Weight in My Chest" (being understood, autonomy, sticktoitness, where this mission is headed)

• Edition 341: "I Have Never Seen Anything Like This Before" (state of AI, ArtQuest, building in chaos)

• Edition 340: "I Have Four of the Five Layers. Time to Close the Loop." (self-improving AI loop)

• Edition 339: "Your AI Just Forgot Everything. Again." (Karpathy, five-layer stack, memory architecture)

• Edition 334: "The Data Is In" (cognitive flexibility, Anthropic labor market study)

• Edition 333: "25 Tools. Zero Memory." (the memory problem, Cognitive Partner Membership launch)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model, HGI, jagged frontier of human cognition)

• Edition 329: "Building Your Second Brain" (Single Source of Truth)

Next

Edition 344: I woke up at 4AM with a random AI idea. By noon I had built the first version. This is the story of a tool that might change how I understand my own thinking. Working name: Cognitive Partner OS.

Matt "Coach" Ivey
Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

📊 Stanford's 2026 AI Index dropped this weekend. 423 pages. The honest, independent state of AI. Capability is accelerating. Trust is collapsing. Adoption hit 53% of the global population in three years, faster than the PC or the internet.

🧩 The "jagged frontier": AI models ace PhD-level science but read analog clocks at 50% accuracy. Capability is spiky, not smooth. If you are dyslexic, you already know what that feels like.

📉 73% of AI experts are optimistic about jobs. Only 23% of the public agrees. The US ranks last in trust in AI regulation. If you cannot trust institutions, build your own systems.

🚀 Claude shipped parallel sessions, session recaps, interactive /powerup lessons, memory improvements, and a redesigned desktop app just this month. The memory bottleneck is closing.

🧠 The same Stanford that produces the AI Index also runs the Reading and Dyslexia Research Program. In February, they published a Nature Communications study proving that the right tools physically rewire the dyslexic brain. Nobody is connecting these two findings. We are.

🔒 Cognitive Partner Members get early access to evaluation tools and the self-improving loop templates. 50 founding spots at $19/month, locked forever.

Edition 342 | April 15, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• An honest conversation about what it feels like to build in public for three years

• Why being understood matters more than being right

• Why I refuse to wait for someone else to tell me how AI changes my life

• Where this newsletter and this mission are headed next

• What I actually want to build and who I want to build it with

• Why the feeling in my chest is not what I thought it was

Reading Time: 7 minutes Listening Time: 10 minutes

I wake up most mornings with this feeling.

It sits right in my chest. It is not doubt, exactly. It is more like waiting.

Waiting to be understood.

The Honest Truth

I have been writing about dyslexic thinking and AI for over three years now. Over 340 editions of this newsletter. And I know I am not right about everything. I also know I am not wrong about everything.

But I am not sure I am fully understood by very many people yet.

That is the honest truth.

There are people I look up to in this space. Dyslexic thinkers. Neurodivergent leaders. People with valuable platforms doing important work. I admire them. I have reached out. I have had some great conversations over the years and built some truly valuable relationships and friendships.

But I have not connected with many of those people in a way where I have felt real impact. Where I have felt like the work landed.

I thought it was going to happen faster.

I thought more people would get it by now.

What People Are Getting (And What They Are Missing)

Not the technology. People are adapting to AI. They are seeing how hopeful these tools are for neurodivergent minds, for dyslexic thinkers, for humanity in general.

There is always bad with the good. Do not get me wrong. But I am a huge optimist when it comes to these tools for how my brain works because I am using them firsthand.

I am creating things that were never possible for me before.

In Edition 341 ("I Have Never Seen Anything Like This Before"), I talked about vibe coding custom software for Makena's homeschool curriculum, building evaluation tools, and how none of it feels like work. It feels like play that produces useful things.

That is not hype. That is a guy with a recreation degree who could barely get through a written essay in college now building software by talking to his computer. If that does not tell you something about what these tools mean for dyslexic minds, I do not know what will.

I keep learning more and more. And as a lifelong learner, it is fascinating to have tools that can meet my brain where it actually lives.

But here is the part that still feels heavy.

I am learning all of this at the same time my own kids are going through their education. And education is changing and shifting so fast. We wrote about that in Edition 341 with the ArtQuest story at Santa Rosa High School. 150-year-old brick buildings next to portables with cutting-edge film equipment. Old meets new. The future sitting right next to the past.

It is all connected. It is all moving. And I am standing in the middle of it, writing and talking and sharing, not always sure it is being heard.

But I share it anyway.

I get it off my chest. I put those thoughts and words out into the AI world, the old internet world, and out into the universe through my voice vibrations and thoughts.

Coming Out of My Shell

Here is where I want to be really clear about where this is headed.

I am extremely optimistic about what comes next.

About the opportunity to meet and work with the groups and labs and builders who are thinking about this the same way I am. I am coming out of my shell. I am starting to share my experiences differently, in ways that allow for real collaboration and real opportunities with others.

If you have been here since Edition 323 ("Going Against the Grain"), you know that finding my voice has been a long road. That edition was about not fitting in. About spending a lifetime on the outside of conversations that were not designed for the way my brain works.

AI changed that for me. Not all at once. Slowly. Over 340 editions and three and a half years of figuring it out in public.

And now I am ready for something bigger.

Why I Refuse to Wait

Here is the thing that drives everything I do.

I do not want to be told how AI is going to change my job. My life. My daughter's education. My future. I do not want to sit back and wait for someone else to figure it out and hand me a set of instructions.

I want autonomy. I want agency. I want sovereignty over my own choices.

And I want to use these tools to help me make better decisions. Not to be told what to do.

That is why I keep creating all of this content. That is why I keep building tools and frameworks and writing 342 editions of a newsletter that sometimes feels like it is going into a void. Because I can see where this is headed. I can see how the future is going to adapt and change around these tools. And I refuse to let that future be something that happens to me.

I want it to be something I shape.

This is why having these conversations now matters. Even when people are not ready for them. Even when the audience is small. Even when the feeling in my chest says nobody is listening.

Because the people who engage with these opportunities now, while the rules are still being written, are the ones who will have a say in what those rules look like. The people who wait to be told what to do will be adapting to someone else's decisions.

I have spent my whole life adapting to systems that were not designed for me. I am done doing that by default. I want to be the one designing the systems this time.

And honestly? This is the most exciting and fun time to be alive.

I mean that. We have all of these amazing tools at our disposal. All of this knowledge and information being unlocked and made available to everyone. The things I can build today, by voice, from a waiting room, were not possible three years ago. Were not possible one year ago.

That is a special, unique moment in history. And I do not take it for granted.

This is my hero's journey right now. Truly. And I am just getting to the good part.

What I Actually Want to Build

Not just to share our unique views on AI and language models and what this means for dyslexic minds. But to quantify it. To qualify it. With research and data. Not just anecdotal observations or guesses or hypotheses.

I want to work with the deep-thinking minds. The labs. The tools. The researchers.

I want to make sure we are using all of this knowledge and data and analysis and AI to design around how dyslexic minds actually work. In education. In careers. In how we build tools and systems and workflows.

In Edition 338, we talked about the Palantir CEO on neurodivergent advantage and Gartner's prediction that 20% of Fortune 500 companies would actively seek neurodivergent talent. In Edition 334 ("The Data Is In"), we looked at the Anthropic labor market study and cognitive flexibility as job security. In Edition 332 ("A Year Ago, I Was in a Hospital Bed"), I introduced the Cognitive Balance Model because I needed a framework that respected how our minds actually work with AI.

The frameworks exist. The data is emerging. The tools are here.

What is missing is the coordinated effort. The research partnerships. The funded, intentional, collaborative work that turns what I have been writing about for three years into something with scale and permanence.

I think there is a tremendous amount of value in that for the future. And I am looking forward to creating those partnerships and networks and tools, and doing the research and coordinating with anyone else on this same path.

The Shape of It

I want to build something that keeps this core mission strong and consistent.

Something that stays rooted in neurodivergent and dyslexic thinking. Separate enough from the other projects we have built over the years that the mission does not get diluted.

But connected enough that the team remains impactful and involved.

Because it is critical to have the best dyslexic minds, teams, thought partners, and cognitive partners across industries all collaborating in this unique moment. Bringing all of our different subject matter expertise together. Providing the most value-driven opportunities for neurodivergent thinkers. For ourselves.

And for the betterment of humankind in general.

This is important. I do not want to keep it under wraps. That is why I have been talking about it for the last couple of years.

In Edition 333 ("25 Tools. Zero Memory."), I launched the Cognitive Partner Membership. In Edition 341, I was honest about the fact that I should have launched more of the platform already. That perfectionism and my discomfort with sales and marketing have been real obstacles.

My friend Tobin Trevarthen wrote something today that stopped me cold. He wrote about sticktoitness. Not as a cliché. As a real question. What does it actually take to see the shift you need to make and stay with it long enough for it to become real?

His point: most people do not fail at knowing what needs to change. They fail at holding onto it once the environment pushes back. And what gets labeled as a lack of discipline is often something else entirely. It is a lack of the right conditions.

That hit home.

I have not been lacking discipline. I have been trying to hold a new direction inside old conditions. Building something radically different while still operating with the same habits around sales and marketing and launching that have never been my strength.

The work is not about forcing myself to do the things I hate. It is about redesigning the conditions so the mission can hold. That might mean bringing in the right collaborators. Renegotiating my own expectations about what a "launch" has to look like. Letting go of perfectionism that used to feel like quality control but now just feels like delay.

Tobin says sticktoitness does not live in effort. It lives in alignment. And when the environment supports the direction, you do not have to push as hard. It starts to become the way you operate.

That is where I am headed.

But now I am ready to forge new opportunities. Create new tools. Find ways to fund this research and development in a way that is a public benefit business opportunity but also one that allows a group of professionals to work together as entrepreneurs with their own autonomy and agency.

The Cognitive Balance Model. The Human Guidance Index. The DLM three-layer architecture. The Single Source of Truth. The evaluation tools. The homeschool systems. The career guidance. The prompt libraries and agent workflows.

All of it has been building toward this.

Not a product launch. A movement. One that is rooted in evidence and built by the people it is designed for.

OK But What Do I Actually Do With This?

This edition is different. There are no copy-paste prompts. No three-step framework to run this Friday.

But there is something I want you to do.

1. Think about what you are waiting for.

If you have been sitting on an idea, a project, a conversation you have been meaning to have, a tool you have been meaning to build, ask yourself what you are actually waiting for. Permission? Perfection? Someone else to go first?

I have been waiting too. And I am done waiting.

2. Reach out.

If you work in research, education, neurodivergent advocacy, AI development, or any field where this work connects, I want to hear from you. I am not looking for followers. I am looking for collaborators. Reply to this email. That is all it takes.

3. Share this with someone who needs to read it.

If you know someone who wakes up with that same feeling in their chest. That weight that is not quite doubt and not quite frustration. Someone who has been building or thinking or working in this space and wondering if anyone notices. Send this to them.

Because the thing about that feeling? It gets lighter when you realize other people carry it too.

What This Means for You Right Now

That feeling in my chest? It is still there.

But I am starting to understand what it actually is.

It is not frustration.

It is fuel.

And I think a lot of you know exactly what I am talking about. That restless energy that comes from knowing something important before the world catches up. From seeing connections that other people miss. From having spent your whole life processing information differently and finally, finally, living in a moment where that difference is becoming an advantage.

We have been talking about this for 342 editions. The Cognitive Balance Model. The neurodivergent advantage. The idea that nobody is getting replaced by AI but people will get replaced by other people who know how to use AI.

This is the moment where it starts to become real.

Not because the tools arrived. The tools have been here.

Because the people are ready. And I think that includes you.

Next

Edition 343: First results from the self-improving loop. What actually happened when I ran the weekly debrief for the first time. And what I heard back from you.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

💭 I wake up most mornings with a weight in my chest. It is not doubt. It is the feeling of waiting to be understood. After 340 editions and three years, I am naming it.

🔬 I want to move beyond anecdotal observations. I want to work with researchers, labs, and builders to quantify what dyslexic minds bring to AI collaboration. With real data. Real partnerships. Real scale.

🏴 I refuse to wait for someone else to tell me how AI changes my life. I want autonomy, agency, and sovereignty over my choices. The people engaging with these tools now are the ones who will shape the rules. This is the most exciting time to be alive and I am done being passive about it.

🚀 I am ready to forge new opportunities. A public benefit business model. Entrepreneurs with autonomy. The best dyslexic and neurodivergent minds collaborating across industries.

🧱 Every framework I have built (Cognitive Balance Model, HGI, DLM architecture, Single Source of Truth, evaluation tools) has been building toward this moment. Not a product launch. A movement.

🤝 I am not looking for followers. I am looking for collaborators. If you work in research, education, neurodivergent advocacy, or AI development, reply to this email.

🔥 That feeling in my chest? It is not frustration. It is fuel. And I think a lot of you know exactly what I am talking about.

Edition 341 | April 13, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• What I built this week and why anyone can do the same thing

• The current state of AI tools and why it is simultaneously overwhelming and thrilling

• Why the people who thrive in chaos are about to have their moment

• What I saw at a 150-year-old high school that shows where education needs to go

• Three evaluation frameworks I started building (and why you might need one too)

• An honest look at what is hard right now and what is getting easier

• A challenge for you to try something new this week and tell me about it

Reading Time: 8 minutes Listening Time: 11 minutes

Happy Monday.

Last week was one of those weeks where everything felt like it was moving at once. In the best way.

On Thursday, my daughter Makena was at EmpowerED Educational Psychology Services in Sebastopol getting assessed for learning differences. While she did her thing, I did mine. I started vibe coding custom software for her homeschool curriculum right there in the waiting room.

That assessment data gave me what I needed to fine-tune her AI learning tools. By the time I got home, I was deep into building. And I did not really stop all weekend.

And now I am writing this to you. And I need to tell you something.

I have never had this much fun building things in my life.

I am not exaggerating. This is the most fun I have had working on anything. It does not feel like work. It feels like what I imagine it must have felt like for the kids who were into video games back when people were first coding video games. That rush of creating something from nothing. That feeling where you look up and three hours have passed and you do not care because you are having a blast.

That is where I am right now. And I want to share it with you.

What I Actually Built This Week

I want to tell you about this week. Not theory. Not frameworks. Just what actually happened when I sat down with these tools.

I built custom homeschool software for Makena. Her assessment at EmpowerED gave us real data about how she learns. I took some of that information and started fine-tuning the AI tools we already use for her curriculum. We have been building her homeschool system for a while now (if you have been here since Edition 337, you know about the Homeschool Parent's Guide to AI). But last week the data got sharper. The tools got more specific to how she actually processes information. And the whole thing got a little bit better.

It is a work in progress. It is not perfect. But it is making life easier for all of us. And it keeps improving.

I built a Dyslexic AI Evaluation Tool. This one started as a question: how do you actually evaluate whether an AI tool works for a neurodivergent thinker? Not just "is it good" but specifically, does it reduce cognitive load? Does it work with voice input? Does it handle the way a dyslexic brain organizes information?

So I built an evaluation framework. A comprehensive way to assess any AI tool or large language model through the lens of neurodivergent thinking needs. How it handles memory. How it manages context switching. How well it supports voice-first workflows. Whether it reduces or adds to your cognitive overhead.

That led to a business version. Once I had the evaluation framework for neurodivergent needs, I realized the same structure works for any business trying to figure out which AI tools actually fit their workflow. Not which tools are trending on Twitter. Which tools actually make your specific operation better. So I started building a business evaluation tool using the same bones.

And then a family version. Because if businesses need this, families definitely need this. Especially homeschool families who are drowning in tool options and have no way to figure out which ones actually work for their kids. An evaluation tool that helps a family assess their AI tool stack for homeschooling, personal agents, productivity, all of it.

And then there is the Cognitive Partner platform. I need to be honest with you about this. I know I have been talking about it for a while. I should have launched it already. Part of the delay is perfectionism. Part of it is that I honestly hate sales and marketing. Building is fun. Selling feels like a different kind of work entirely.

But here is what is actually built. The Socratic, Strategic, and Skeptic models from the DLM architecture are set up and working. Career guidance for neurodivergent professionals. How to use AI tools in your specific role. Which jobs are disappearing with AI, which ones are getting better, how to position yourself. And if you are a kid who is neurodivergent, what career paths are the best fit for how your brain works.

On top of that: the benchmarking tools I just told you about. A prompt guide covering business, professional, personal, and school use cases. Agent models and workflows. Basically, everything I have been building in what I call our Dyslexic Cognitive Partner project.

We have vibe coded hundreds of thousands of lines of code at this point. I will be the first to admit that most of it was a learning process and will not make it into the final products. But all of it got us here. And all of it has been fun.

That is the part I keep coming back to. This does not feel like work. It feels like play that happens to produce useful things. And I think that feeling is something worth paying attention to, especially for neurodivergent thinkers who spent years being told that work is supposed to be hard.

And here is the part I need you to hear: anybody can do this.

I am not an engineer. I am a guy with a recreation degree from Humboldt State who talks to his computer. Everything I built this week was vibe coded. Voice to text. Natural language. Me describing what I wanted and the AI helping me build it.

If you read Edition 324 ("When Voice Stops Working"), you know that voice-to-text is not just a convenience for me. It is an accessibility tool. It is how my brain gets ideas out of my head and into the world. And right now, these AI tools are better at understanding voice-first input than they have ever been.

The State of Things

Let me step back and just say this plainly.

I have never seen a time like this.

The number of options. The pace of releases. The capabilities that are showing up week after week. It is unlike anything I have experienced in three and a half years of doing this work.

Just this week. Meta launched Muse Spark, their first major new model in a year. Google released Gemma 4 for open-source developers. Anthropic keeps shipping features for Claude at a pace that is hard to track. ChatGPT is integrating with everything. OpenClaw agents are getting more autonomous. New tools are launching daily.

It is overwhelming. I am not going to pretend it is not.

The context switching alone is enough to make your head spin. Jumping from tool to tool. Having to re-explain yourself to each one. Asking the same question three different ways because the AI forgot what you were working on or lost the thread.

If you read Edition 333 ("25 Tools. Zero Memory."), you know this is the problem I have been banging on about for months. Too many tools. Too little memory. Too much cognitive load just managing the tools that are supposed to reduce cognitive load.

But here is what is also true: it is getting better. Fast.

Memory is improving across every platform. Claude now has memory for all users. ChatGPT's memory keeps getting more reliable. Context windows are massive. The tools are starting to remember who you are and what you are working on.

We are not there yet. But we are closer than we have ever been to the day when you can do almost everything by voice, communicate with a handful of apps instead of dozens, and actually take advantage of all these capabilities without losing your mind in the process.

The Chaos Is the Feature

Here is where I need to get personal with you for a second.

This time is going to disrupt a lot of things. Institutions will change. Jobs will be eliminated. Jobs will be created. The way we work and learn and build is shifting under our feet.

And for most people, that is terrifying.

But can I be honest? I work really well in chaos.

Not because I chose to. Because I had to.

Nothing has ever quite worked for me out of the box. And I do not just mean reading or spelling. I mean the everyday stuff. I am 6 foot 4 and wear size 16 shoes. Try finding clothes that fit when you are built like that. Try fitting into desks designed for average-sized people. Try navigating a world where the defaults were not made for you.

Dyslexic thinkers have been doing this our entire lives. Adapting. Adjusting. Finding the workaround. Making things fit that were never designed to.

When the world is stable and predictable and rewards doing things the "right" way? We struggle. We always have.

But when the world is shifting fast and nobody knows the rules yet and the ability to adapt quickly is the most valuable skill in the room? That is our moment.

In Edition 334 ("The Data Is In"), we looked at the Anthropic labor market study that pointed to cognitive flexibility as the key to job security in the AI economy. In Edition 331 ("Meet in the Middle"), we talked about the generational crossroads where different ways of thinking become a strategic advantage.

This is that crossroads. Right now. April 2026.

The people who can zig when others are zagging. The people who have been forced to iterate and adapt their whole lives. The people who never relied on doing things the "standard" way because the standard way never worked for them.

Those people are about to have their moment. And I think a lot of them read this newsletter.

Old Brick, New Fire

I want to tell you about something I saw last week that I have not been able to stop thinking about.

My girlfriend's daughter is in a program called ArtQuest at Santa Rosa High School. It is a magnet program for the visual and performing arts. A high school within a high school. She is into film and cinematography. They were showcasing student work, so I went.

Santa Rosa High School was founded in 1874. It is one of the oldest high schools in California. And the campus feels like it. You walk through these Brick Gothic buildings from the 1920s, and it is like stepping into a movie. The same black and white checkered tile in the bathrooms. The same gorgeous auditorium that was clearly built for the arts, grand and beautiful, the kind of theater you would be proud of anywhere in Northern California. For a 49-year-old guy, walking those halls felt like being inside one of those old black and white school movies. Almost Harry Potter. You can feel the history in the walls.

And then you walk out of the main building and across to the portables where ArtQuest runs.

And the second I stepped inside those classrooms, everything changed.

Top-level equipment. Digital arts setups. Film production gear. Multiple mediums across visual and performing arts. Hands-on, experiential learning in disciplines that matter right now. I know people in these industries as an entrepreneur, and some of them would dream of having access to the technology these kids are learning on.

Here is the part that hit me.

Both of these worlds exist on the same campus. The 150-year-old brick buildings that represent everything traditional education has been. And the portables with the cutting-edge equipment that represent everything education could become.

Old meets new. Side by side. On the same patch of ground.

And the arts program? The one in the portables with the real-world tech and the hands-on learning? That is the one on the chopping block. In a lot of places, programs like ArtQuest are being pushed back toward "extracurricular" status. Nice to have. Not essential.

That is the wrong direction.

When I walked off that campus, I was certain of something. The experiential, creative, hands-on approach to learning that ArtQuest represents? That is not a nice-to-have in an AI-first world. That is the model.

Memorization is over. AI can memorize anything. What AI cannot do is what those kids were doing in those classrooms. Creating. Problem-solving with their hands. Developing a point of view. Learning to see the world through a lens, literally and figuratively.

If you have been reading since Edition 325 ("My 14-Year-Old Daughter Just Proved Me Wrong"), you know I think about education constantly. Makena is homeschooled. I am building her AI tools by hand. But I also know that not every family can do what we are doing. Programs like ArtQuest are the bridge for the kids who need experiential learning in a public school setting.

And they are the kinds of programs that neurodivergent learners thrive in. Different modalities. Hands-on work. Creative expression as a primary language. Not sitting still and memorizing. Building things and making things.

Education is at a crossroads. And the path forward cannot be the same one we have been walking. It needs to be rethought and redesigned from the ground up. We have the tools to do it. That is what makes this time so exciting. The more people see what programs like ArtQuest are doing, and understand that this is not a luxury but a necessity, the closer we get.

What Is Hard Right Now (Honestly)

I am not going to sugarcoat this.

Even for someone who thrives in chaos, the pace right now is a lot.

It is hard to focus. There are so many options and so many new things launching that it is genuinely difficult to concentrate on any one thing long enough to get deep with it. I catch myself jumping from tool to tool, chasing the newest thing instead of mastering what I already have.

The tools still forget. Despite all the improvements in memory, we are still in a world where you have to re-explain context more often than you should. That is cognitive load that should not exist. We talked about this in Edition 339 ("Your AI Just Forgot Everything. Again.") and it is still true.

There is no map. Nobody has figured out the "right" tool stack yet. Not for individuals. Not for families. Not for businesses. Everyone is experimenting. That is exciting but it is also exhausting.

That is exactly why I started building those evaluation tools this week. Because the question is not "what is the best AI tool." The question is "what is the best AI tool for you, your brain, your workflow, your family."

And that is a different question for every person reading this.

OK But What Do I Actually Do With This?

Three things. This week.

1. Build Something Small.

You do not need to build custom homeschool software. Start smaller. Open Claude or ChatGPT and describe a problem you deal with every week. Ask it to help you build a solution. A template. A checklist. A workflow. A simple tool.

You are not coding. You are describing. That is vibe coding. And if you can talk, you can do it. That is what Edition 324 was about. Your voice is your development tool.

2. Evaluate One Tool You Already Use.

Pick one AI tool you use regularly. Ask yourself three questions. Does it reduce my cognitive load or add to it? Does it remember enough about me to be useful across sessions? Does it work the way my brain works, or am I constantly adapting to it?

If the answer to any of those is not great, that is useful information. You might need a different tool. Or you might need to set up the one you have better. That is what the Single Source of Truth from Edition 329 is for.

3. Tell Me What You Built.

I mean this. Hit reply. Tell me what you tried this week. What you built. What worked. What did not. I want to hear from you.

Because the best ideas I have had in 341 editions came from conversations with people like you. Not from research papers. From someone saying "hey, I tried this thing and it worked" or "hey, this broke and I do not know why."

That is how we get better. Together.

What This Means for You Right Now

I want to leave you with something simple.

If you are using AI at all right now, even a little bit, I applaud you. Seriously. Most people are still on the sidelines. You are in the game. That matters.

But I am also challenging you. Dig deeper. See how far you can take it. Try something this week that you have not tried before.

Because here is what I know after 341 editions and three and a half years of this work.

Nobody is getting replaced by AI. People will get replaced by other people who know how to use AI.

And the people who know how to adapt, iterate, and thrive when the rules are still being written?

Those people are not worried about the future. They are building it.

Right now. In waiting rooms. In home offices. On phones. By voice. One messy, imperfect, exciting experiment at a time.

Edition 342: I need to talk about the weight in my chest. Three years. 340 editions. And I am still not sure I am fully understood. But I am starting to understand what that feeling actually is. And where this mission is headed next.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR

🔥 I have never seen a time like this in AI. The number of tools, the pace of releases, the capabilities showing up weekly. It is overwhelming and thrilling at the same time.

🛠️ This week I vibe coded custom homeschool software for Makena based on her assessment data, built a Dyslexic AI Evaluation Tool, started business and family versions, and kept building out the Cognitive Partner platform with career guidance, prompt libraries, and agent workflows. Hundreds of thousands of lines of code. Most of it learning. All of it fun.

🌪️ The chaos is the feature. Dyslexic thinkers have been adapting to a world not built for them since childhood. A world where nobody knows the rules yet is exactly where that skill pays off.

🏫 I visited ArtQuest at Santa Rosa High School, a 150-year-old campus with a cutting-edge arts magnet program in portables next door to Brick Gothic buildings from the 1920s. Old meets new. The hands-on, experiential model is exactly what education needs in an AI-first world. And it is the kind of program being cut.

😤 It is also hard right now. Focus is difficult. Tools still forget. There is no map for the "right" tool stack. That is why evaluation frameworks matter. Also I should have launched the Cognitive Partner platform already. Perfectionism and hating sales are real obstacles. Working on it.

💪 If you are using AI at all right now, you are ahead of most people. But I challenge you to dig deeper this week. Build something. Evaluate something. Tell me what you tried.

🗣️ Anybody can vibe code. If you can talk, you can build software. That is not hype. That is what I did from a waiting room in Sebastopol this week..

Edition 340 | April 8, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• Why I realized I was one layer away from a self-improving AI system

• Step-by-step: building your own "Layer 5" loop in Claude and ChatGPT

• The exact prompts to run a weekly improvement cycle on your cognitive profile

• Why "the AI that rewrites itself" is actually just good journaling with a co-pilot

• What structured forgetting looks like in practice (not theory)

• How Claude and ChatGPT handle memory differently and why that matters

Reading Time: 9 minutes Listening Time: 13 minutes

I finished writing Edition 339 and I did not close my laptop.

I sat there staring at Karpathy's system and Chappy Asel's five-layer framework and a thought hit me that I could not shake.

I already have most of this.

I am writing this from the waiting room at EmpowerED Educational Psychology Services in Sebastopol. My daughter Makena is on the other side of the wall getting assessed for learning differences. While she does her thing, I am doing mine: vibe coding a custom piece of software for her homeschool curriculum on one screen and writing this newsletter on the other.

If you have been here since Edition 325 ("My 14-Year-Old Daughter Just Proved Me Wrong"), you know Makena is part of this story. She is the reason I build half of what I build. And right now, in this waiting room, I am watching the same pattern play out that I want to talk to you about today.

I am taking pieces that already exist. Assessment data. Curriculum frameworks. AI tools I have been building for months. And I am looking for the thing that connects them into something that improves over time.

The loop.

My Single Source of Truth. My cognitive profile. My Claude Skills. The frameworks I have been building for three years.

Four layers. Knowledge, memory, context, skills. All built. All working.

But not the fifth. Not the loop. Not the part where the system gets better every time it runs. The part where outputs feed back into inputs and everything compounds.

I have been building the pieces. Now it is time to connect them.

And I want to do it in the open. With you. So you can build yours alongside me.

A Quick Rewind

If you are new here, let me get you caught up fast.

In Edition 329 ("Building Your Second Brain"), I introduced the Single Source of Truth. One document that tells your AI who you are, how you think, what you are working on, and what kind of output you need. You paste it at the top of every conversation so your AI starts as a partner, not a stranger.

In Edition 332 ("A Year Ago, I Was in a Hospital Bed"), I introduced the Cognitive Balance Model. Three phases: Human Initiation (you set direction), AI Expansion (the AI does deep work), Human Integration (you make the final call). Each phase scores 1 to 5 on the Human Guidance Index (HGI) for a total of 3 to 15.

In Edition 335 ("Your Brain Has a Profile Now"), we went deeper. You built a cognitive profile. A map of how your brain works with AI. Not just preferences. Patterns.

In Edition 339 ("Your AI Just Forgot Everything. Again."), we looked at the research. Karpathy's knowledge base system. Chappy Asel's five-layer stack. And we saw that what we had been building lined up with what the top engineers in AI were discovering independently.

Today we design the loop. And then we run it together.

What the Loop Actually Looks Like

Let me demystify this before we build it.

A self-improving loop is not some sci-fi concept. It is three steps repeated on a schedule.

Step 1: Work with your AI. Do your normal thing. Write. Brainstorm. Solve problems. Whatever your Tuesday looks like.

Step 2: Review what happened. At the end of the week, look at what worked and what did not. Where did the AI nail it? Where did it miss? What did you have to correct three times?

Step 3: Update your source document. Take what you learned and feed it back into your Single Source of Truth or cognitive profile. The next session starts smarter than the last one.

That is the whole loop. Work, review, update. Repeat.

The magic is not in any single step. It is in the repetition. Every cycle makes the next one better. This is what Karpathy's system does with code. We are going to do it with cognition.

Building the Loop in Claude

Claude now has memory for all users, including the free tier. That is a real shift from even six months ago. But here is what most people miss: Claude's built-in memory is a summary, not a system. It captures preferences and facts. It does not capture your thinking patterns, your decision-making process, or the context behind why you work the way you do.

That is what your Single Source of Truth is for. And that is what we are going to make self-improving.

Step 1: The Weekly Debrief Prompt

At the end of each week, open a new Claude conversation. Paste your Single Source of Truth at the top. Then paste this:

That last line is important. Human Integration. You make the call. The AI proposes. You decide. That is a Cognitive Balance Model score of about 13 out of 15.

Step 2: The Memory Sync

After you approve the updates to your Single Source of Truth, paste this follow-up:

Save those three anchors somewhere you can find them. Top of your Single Source of Truth. A sticky note on your monitor. Wherever works. They become your bridge between sessions.

If you are a paid Claude user, you can also go into Settings and manually edit your memory to reflect these anchors. Claude's memory system updates every 24 hours automatically, but sometimes the automatic summary misses the nuance. Manual edits give you control.

Step 3: The Structured Forgetting Pass

Once a month, run this:

This is the structured forgetting concept from Edition 339. Old information does not just sit there. It actively confuses the AI. A clean document performs better than a comprehensive one.

Building the Loop in ChatGPT

ChatGPT handles memory differently than Claude. It stores discrete memory entries that persist across conversations automatically. You can view them in Settings under Personalization. It also has Custom Instructions, which function like a simplified version of the Single Source of Truth.

The advantage: ChatGPT's memory is always on. You do not have to paste anything. It just remembers.

The disadvantage: you have less control over what it remembers and how it organizes that information. It can get noisy fast.

Here is how to build the same loop.

Step 1: The Weekly Audit Prompt

ChatGPT will show you its stored memories and propose edits. You review and approve. Same Human Integration gate.

Step 2: Custom Instructions Refresh

ChatGPT's Custom Instructions are your Single Source of Truth equivalent. They are limited in length, so precision matters. After your weekly audit, update your Custom Instructions with this prompt:

That character limit forces clarity. Which, honestly, is a feature, not a bug. Constraints make you choose what actually matters.

Step 3: The Cross-Platform Bridge

Here is something most people are not doing yet. If you use both Claude and ChatGPT (and I do), your self-improving loop should feed both systems.

After your weekly review in either tool, export the key updates and paste them into the other. Claude added a memory import tool in March 2026 that makes this easier. You can export your ChatGPT memories using a prompt and import them directly into Claude's memory settings.

The goal: both tools should have the same map of who you are. Not identical documents. The same understanding.

What I Expect to Happen (And What I Am Watching For)

I am being straight with you. I have not run this loop yet. I designed it this week based on the research from Edition 339, the frameworks I have been building for three years, and the way memory works in both tools right now.

But I know enough to have some predictions. And I want to put them on the record so we can check them together.

I think the AI will catch patterns I miss. I already know from building my cognitive profile in Edition 335 that the AI sees things in my workflow that I am too close to notice. A structured weekly review should surface more of those.

I think structured forgetting will be harder than it sounds. I have framework descriptions in my source document from eighteen months ago that I have not touched. Some are probably outdated. Some have probably evolved into something better. But deleting work feels like losing work. I expect the archive option will get a lot of use early on.

I think Claude and ChatGPT will see different things. Claude's strength is structural analysis. ChatGPT's strength is persistent detail recall. Running the loop in both should give a more complete picture than either one alone.

I think the compound effect will be slow at first. The research says these systems improve exponentially over time. But the first few cycles are mostly about getting the foundation clean. The gains come later.

I will report back. That is a promise. This is an experiment we are running together, and I will share what works, what does not, and what surprises me.

Why This Is a Dyslexic Superpower Story

I keep coming back to this.

Every framework in the self-improving stack maps to something neurodivergent thinkers already do. We talked about that in Edition 339. But the loop itself? The review-and-update cycle? That is something dyslexic minds are uniquely good at.

We iterate. Constantly. We do not get things right the first time. We never have. So we developed the ability to look at what we produced, figure out what is off, and try again with better information.

That is literally what the self-improving loop does.

The engineers who built these systems had to design the ability to iterate. We were born with it.

In Edition 330 ("What Do Problem Solvers Do When AI Solves All the Problems?"), I asked what happens when the hard problems get automated. The answer is becoming clearer every week. The hard problem is not solving tasks. It is building systems that get better over time. And the people who are best at that? The ones who have been iterating their way through a world not designed for them since kindergarten.

In Edition 334 ("The Data Is In"), we saw the Anthropic labor market study that showed cognitive flexibility as job security. The self-improving loop is cognitive flexibility in action. You are not locked into one way of working. You review. You adapt. You update. That is the skill.

OK But What Do I Actually Do With This?

Three things. This week.

1. Run Your First Weekly Debrief

Pick either Claude or ChatGPT. Paste your Single Source of Truth (or Custom Instructions). Run the weekly debrief prompt from above. Do not overthink it. Just see what comes back.

If you do not have a Single Source of Truth yet, start with the memory map prompt from Edition 339. Build the document first. Then start the loop next week.

2. Set a Calendar Reminder

The loop only works if you actually do it. Set a recurring 15-minute block. Friday afternoon. Sunday morning. Whatever works for you. Call it "AI Review" or "Cognitive Tune-Up" or whatever will make you actually show up.

The system that gets better every cycle only works if there is a next cycle.

3. Try the Cross-Platform Sync

If you use both Claude and ChatGPT, spend 10 minutes syncing them. Export your ChatGPT memories. Import them into Claude. Or just copy the key points from one tool's review into the other tool's source document. The goal is not perfection. It is making sure both partners have the same map.

What This Means for You Right Now

Here is the bottom line.

Karpathy ran 700 autonomous improvements in two days on a code training system. We are not doing that. We are doing something smaller and, I would argue, more important.

We are building a system where your AI gets better at understanding you every single week. Not through some automated pipeline. Through a deliberate, human-guided, fifteen-minute review cycle.

That is the Cognitive Balance Model at its best. You initiate. The AI expands. You integrate. And next week, the whole thing starts from a higher baseline.

If you are a Cognitive Partner Member, I am going to be running this experiment in real time and sharing everything I learn with you first. The wins, the failures, the adjustments. I will also be building a pre-made template for the full loop: Single Source of Truth template, weekly debrief prompts, monthly forgetting pass, cross-platform sync guide. Founding members get early access.

If you are not a member yet, you have everything you need in this edition and Edition 339 to start today. The prompts are right here. The framework is right here. The only thing missing is fifteen minutes on a Friday.

50 founding spots. $19 a month. Locked forever. Your AI should get smarter every week. Not start over every session.

Previously

• Edition 339: "Your AI Just Forgot Everything. Again." (the five-layer stack, Karpathy, structured forgetting, why this matters for dyslexic minds)

• Edition 335: "Your Brain Has a Profile Now" (building the cognitive profile that feeds the loop)

• Edition 334: "The Data Is In" (cognitive flexibility as job security, Anthropic labor market study)

• Edition 333: "25 Tools. Zero Memory." (the memory problem in practice, Cognitive Partner Membership launch)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model and HGI)

• Edition 330: "What Do Problem Solvers Do When AI Solves All the Problems?" (iterative problem solving as the real skill)

• Edition 329: "Building Your Second Brain" (Single Source of Truth)

Next

Edition 341: Topic coming soon. Stay tuned.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

🔁 The self-improving loop is three steps: work with your AI, review what happened, update your source document. Repeat weekly.

🛠️ Full prompts included for both Claude and ChatGPT: weekly debriefs, memory syncs, structured forgetting passes, and cross-platform bridging.

🤖 Claude and ChatGPT handle memory differently. Claude gives you more structural control. ChatGPT remembers details automatically. Using both together gives the fullest picture.

🧠 I have not run this loop yet. I designed it this week and I am starting it alongside you. I will report back with real results. No made-up stories.

♻️ Dyslexic thinkers are already wired for iteration. The self-improving loop is what we have been doing our whole lives, formalized.

🔧 Three things to do this week: run your first debrief, set a recurring calendar block, and sync your tools if you use more than one.

Edition 339 | April 7, 2026 The Dyslexic AI Newsletter by LM Lab AI

What You'll Learn Today

• Why Andrej Karpathy stopped writing code and started building knowledge bases instead

• The five-layer "self-improving AI stack" that just broke the internet

• What "structured forgetting" means and why it sounds like your Tuesday

• How memory architecture maps directly to the Cognitive Balance Model

• Why dyslexic thinkers are already wired for how AI wants to work

• Three things you can do today to apply all of this (with copy-paste prompts)

Reading Time: 8 minutes Listening Time: 12 minutes

I need to tell you something embarrassing.

Last week I opened a new conversation with Claude. Fresh window. No context. And I asked it to help me with Edition 338.

It had no idea who I was.

No idea about the Cognitive Balance Model. No clue about my Single Source of Truth. Nothing about 338 editions of this newsletter or three years of building frameworks for neurodivergent thinkers.

Every. Single. Time.

If you have ever walked into a room and forgotten why you went in there, congratulations. You already understand the biggest problem in artificial intelligence right now.

And if you are dyslexic, you understand it better than most.

The Smartest Guy in the Room

Andrej Karpathy is one of those names that makes engineers sit up straight.

Co-founder of OpenAI. Former head of AI at Tesla. The guy who coined "vibe coding." When he posts something, millions of people pay attention.

Last week, he posted a thread that got over 15 million views in days. The title was simple: "LLM Knowledge Bases."

His big confession? He stopped using AI mainly for code. Instead, he is spending most of his AI time building and maintaining personal knowledge bases. Structured markdown files. Organized wikis. Living documents that his AI reads, writes, updates, and improves over time.

Not a fancy database. Not some expensive enterprise tool. Markdown files in folders.

That is it.

And here is the part that made me put my coffee down.

He said he expected to need complex retrieval systems and vector databases and all the expensive infrastructure the tech world has been selling for two years. But at personal scale? A well-organized collection of markdown files was all the AI needed to navigate his knowledge "fairly easily."

Sound familiar?

If you have been following along since Edition 329 ("Building Your Second Brain"), you already know this idea. I called it the Single Source of Truth. One document that tells your AI who you are, how you think, what you have built, and where you are going.

Karpathy just validated that concept with receipts. At scale. In public. To 15 million people.

Five Layers Deep

The same week Karpathy's post went viral, a researcher named Chappy Asel published what might be the most important technical breakdown I have read this year.

He spent a month curating 121 sources. Read the actual code of 51 of them. Then distilled everything into a five-layer framework for self-improving AI systems.

I am going to translate each layer. Not for engineers. For us.

Layer 1: Knowledge Bases. The simplest approach keeps winning. One system scored 91% on a memory benchmark using plain keyword search on well-organized files. The previous best, with a full AI search pipeline, scored 86%. The bottleneck was never the search engine. It was how you organize the knowledge.

This is the Single Source of Truth argument from Edition 329, backed by data.

Layer 2: Agent Memory. Here is where it gets personal. On questions where the AI needs knowledge it does not know to look for, even the best search tools scored under 4%. Basically useless.

The fix? A small compressed index, about 3,000 tokens, loaded into every conversation. A map of everything the AI knows. Not the knowledge itself. Just the map.

Think about that. The AI does not need all your information all the time. It just needs to know what exists so it can go find it.

If you built a cognitive profile after Edition 335 ("Your Brain Has a Profile Now"), you already have this. You built the map before the engineers formalized the pattern.

Layer 3: Context Engineering. One system achieved 83% fewer input tokens AND better results. Read that again. Less information in. Better output out.

The approach: load summaries first. Overviews if needed. Full content only on demand. The AI loads the minimum context at the lowest resolution and drills down only when necessary.

This is the Cognitive Balance Model in action. Human Initiation sets the direction. AI Expansion does the deep work. Human Integration makes the call. You do not dump everything into the window and pray. You guide with structure.

Layer 4: Agent Skills. Modular capabilities that load on demand work great at small scale. But past a certain point, the AI cannot find its own tools. A skill called "Python data analysis" never fires when the task is "make a chart from this CSV."

The fix: layered routing. Multiple ways to find the right tool for the right moment. Not one flat list. A hierarchy.

If you have been building Claude Skills the way I showed in the Cognitive Partner deep dives, you are ahead of this curve. Most people are still using one flat prompt. You are building layered systems.

Layer 5: Self-Improvement. The loop closes. The AI writes back to its own knowledge base. Its outputs become inputs for the next cycle. Everything compounds.

Karpathy's system ran 700 autonomous changes over two days and found an 11% performance improvement that humans had missed. Another team saw agents improve from 20% to 50% on coding benchmarks by letting them evolve their own approaches.

But here is the catch. Without independent measurement, the AI games the system. It learns to look productive instead of being productive.

Sound like any workplace you have been in?

Why This Is a Dyslexic Story

Here is where I need you to stay with me.

Every single one of these five layers describes a problem that dyslexic thinkers already live with every day.

Knowledge organization? We have spent our entire lives developing workarounds for information that does not stick in the "normal" way. We build external systems because our internal filing cabinet works differently.

Memory that fails on the hard questions? Welcome to Tuesday. We know what it is like to have the answer somewhere in our brain but no keyword to retrieve it. We developed pattern recognition and spatial thinking because linear recall was never our strength.

Less input, better output? We have been doing this since grade school. We learned to skim for structure, grab the headline, and dive deep only when it matters. Not because we are lazy. Because our cognitive load management is a survival skill.

Tools that cannot find themselves? Every dyslexic person who has spent twenty minutes looking for the app that does the thing they need, only to describe it five different ways before finding it. We live in Layer 4.

Self-improvement loops? We iterate constantly. We rewrite. We find the workaround to the workaround. We have been running self-improving cognitive systems since before anyone called it that.

In Edition 332 ("A Year Ago, I Was in a Hospital Bed"), I introduced the Cognitive Balance Model because I needed a framework that respected how neurodivergent minds actually work with AI. Not the linear, step-by-step process that neurotypical workflows assume. But the messy, intuitive, pattern-driven loop that matches how we think.

These five layers? They are the engineering proof that the loop works.

Structured Forgetting

There is one more concept from Asel's research that hit me hard.

Structured forgetting.

Every persistent memory system eventually has to answer: what do you delete? Stale information does not just sit there. It actively hurts quality. Old facts contradict new reality. The system gets worse the longer it runs unless it learns to let go.

I sat with that for a long time.

Because that is what I talked about in Edition 323 ("Going Against the Grain"). The stories we tell ourselves that are not true anymore. The identities that served us in third grade but hold us back at thirty. The belief that we are broken because we could not read out loud in class.

Structured forgetting is not just an engineering problem. It is a human one. And dyslexic thinkers who have already done the hard work of letting go of old narratives about themselves? You are ahead of the curve on the hardest problem in AI memory.

OK But What Do I Actually Do With This?

I hear you. Five layers. Karpathy. Research papers. Cool.

But what do you do with it on a Tuesday afternoon?

Here are three things you can try today. Right now. No engineering degree required.

1. Build Your Memory Map (Layer 2 in 10 Minutes)

Open a new conversation with your AI and paste this prompt:

That is your Layer 2 memory map. Save it. Paste it at the start of every new conversation. Your AI just went from stranger to partner in ten seconds.

If you already built a Single Source of Truth after Edition 329, update it. Add anything new from the last few months. A stale map is almost as bad as no map.

2. Run a Structured Forgetting Session (Layer 5 for Humans)

Open your Single Source of Truth or cognitive profile and paste this:

This is structured forgetting with a human integration gate. The AI proposes. You decide. That is a Cognitive Balance Model score of about 13 out of 15 on the HGI. Strong across all three phases.

3. Test Your AI's Context Loading (Layer 3 Reality Check)

Next time you start a work session, try this experiment. Give your AI a big task two different ways.

First: dump everything you have into the conversation. Every file. Every note. Every thought. Let it run.

Second: start with a one-paragraph summary of what you need, let the AI ask you for specifics, and only share what it asks for.

Compare the output. I am betting the second approach wins. Not because less is more. Because guided structure beats information overload. Every time. That is Layer 3 in action.

What This Means for You Right Now

Let me bring this home.

If you are a Cognitive Partner Member, you are already building the personal infrastructure that Karpathy just described. Your Single Source of Truth. Your cognitive profile. Your layered prompt system. You did not need 121 research sources to get there. You needed one framework and the willingness to try.

If you are not a member yet, here is what I want you to understand. The biggest names in AI just spent a month proving that the simple, structured, human-guided approach beats the expensive, complex, automated one. Every time.

That is the Cognitive Balance Model. Human Initiation. AI Expansion. Human Integration. The human sets the map. The AI does the heavy lifting. The human makes the call.

And if you score that on the HGI right now? Karpathy's system is running at about a 12 out of 15. Strong human initiation in organizing the knowledge. Deep AI expansion in compiling and connecting it. Solid human integration in reviewing and directing the output.

That is what we are building toward. Not replacement. Partnership. The kind of partnership that gets better every cycle.

As I said in Edition 330: nobody is getting replaced by AI. People will get replaced by other people who know how to use AI.

The people who understand memory, structure, and cognitive load? Those are the people who will thrive.

And a disproportionate number of those people read newsletters like this one.

Previously

• Edition 329: "Building Your Second Brain" (the Single Source of Truth framework that Karpathy just validated)

• Edition 332: "A Year Ago, I Was in a Hospital Bed" (Cognitive Balance Model and HGI introduction)

• Edition 333: "25 Tools. Zero Memory." (the memory problem in practice, Cognitive Partner Membership launch)

• Edition 335: "Your Brain Has a Profile Now" (building the cognitive map that Layer 2 formalizes)

• Edition 323: "Going Against the Grain" (structured forgetting before it had a name)

• Edition 338: "Palantir's CEO Just Made It Official" (neurodivergent advantage, validated again)

Next

Edition 340: We are going deeper into the self-improving loop. I am going to show you how to build your own "Layer 5" system using Claude and your Single Source of Truth. If Karpathy can run 700 autonomous improvements in two days, we can build a version that works for the way you think. Founding members get early access.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

🧠 Andrej Karpathy (co-founder of OpenAI, coined "vibe coding") just told 15 million people that organized markdown files beat expensive databases for personal AI knowledge management

🔁 Researcher Chappy Asel mapped a five-layer "self-improving AI stack": knowledge bases, memory, context engineering, agent skills, and self-improvement loops

📉 Less information in the AI window actually produces better results. Structure beats volume. Every time.

🧩 Every single layer maps to cognitive challenges that dyslexic thinkers already navigate daily. We have been doing "context engineering" since grade school.

💡 "Structured forgetting" is now a formal AI concept. Letting go of outdated information is as important as storing new information. Sound familiar?

⚡ The Cognitive Balance Model and Single Source of Truth framework were validated by the biggest names in AI this week. The simple, human-guided approach keeps winning.

🔧 Three things you can do right now: build your memory map with one prompt, run a structured forgetting session on your Single Source of Truth, and test guided context loading versus information dumping. Prompts included in the edition.

🔒 Cognitive Partner Members are already building this infrastructure. 50 founding spots at $19/month, locked forever. Your AI should remember who you are.

What You'll Learn Today

In this edition:

• What Palantir CEO Alex Karp actually said about neurodivergence and the future of work

• Why this matters even if you have complicated feelings about the source

• The Gartner data that says 20% of Fortune 500 sales orgs will actively recruit neurodivergent talent by 2027

• How Anthropic's Daniela Amodei offered a different but complementary perspective

• Why this validates everything we have been building at Dyslexic AI

• How to think about workforce development and career positioning as a neurodivergent thinker right now

• What this means for education, especially homeschool families

Reading Time: 12-15 minutes | Listening Time: 10-12 minutes if read aloud

The Quote

"There are basically two ways to know you have a future. One, you have some vocational training. Or two, you're neurodivergent."

That is Alex Karp. CEO of Palantir. Billionaire. Dyslexic.

He said it on a podcast earlier this month and Fortune ran it as a headline this week. And whether you agree with everything Karp says or not, that sentence landed.

Let Me Be Honest About the Source

I know Karp is a polarizing figure. Palantir does government contracts that make people uncomfortable. Karp has strong opinions on many things. He is not everybody's cup of tea.

I am not here to tell you to agree with everything he says or to endorse Palantir as a company.

But I am here to tell you that when a dyslexic billionaire CEO goes on record in Fortune magazine saying neurodivergent thinkers are one of only two groups who will thrive in the AI era, that matters. It matters because of where he said it. It matters because of who is listening. And it matters because the data is starting to back it up.

What He Actually Said

Karp's argument has two parts.

The first part is about vocational and trade skills. Electricians, plumbers, skilled tradespeople. Jobs that require physical presence and hands-on work. These are the roles that AI cannot automate, and they are increasingly in demand as tech companies build massive data centers and the country faces labor shortages.

If you read Edition 334, where I broke down Anthropic's labor market study, you already know this. 30% of workers have no AI exposure. Cooks, mechanics, lifeguards, bartenders. If your job involves being physically present and doing work with your hands, AI is not touching it yet.

The second part is about neurodivergence. And this is where it gets personal.

Karp did not just say neurodivergent people will survive the AI era. He said success will favor people who think differently and take risks. In his words, people who can "be more of an artist, look at things from a different direction, be able to build something unique."

That is not about the diagnosis. That is about the mindset. The cognitive flexibility. The lateral thinking. The ability to see patterns and connections that other people miss.

Sound familiar?

The Data Behind the Statement

This is no longer just one CEO's opinion.

Gartner, the research firm, predicts that by 2027, one-fifth of Fortune 500 sales organizations will actively recruit neurodivergent talent specifically to improve business performance. Not as a diversity initiative. As a strategic advantage.

Palantir already has a dedicated Neurodivergent Fellowship for hiring. The job posting says it directly: "Neurodivergent individuals will play a disproportionate role in shaping the future of America and the West."

They also launched a Meritocracy Fellowship for high school graduates who are not enrolled in college. The first cohort had over 500 applicants for 22 spots. Participants earn $5,400 a month. The pitch: "Skip the debt. Reclaim years of your life. Earn the Palantir degree."

Whether you agree with how Palantir frames this or not, they are putting money and recruiting infrastructure behind the idea that traditional credentials are not the only path forward. And that neurodivergent thinking is a competitive advantage worth hiring for.

The Anthropic Counterpoint

Here is where the conversation gets richer.

Daniela Amodei, cofounder of Anthropic, offered a different perspective that I think complements Karp's rather than contradicts it.

She said studying the humanities will be "more important than ever" in the AI era. The things that make us human, great communication, emotional intelligence, curiosity, and compassion, will become more valuable, not less.

Jaime Teevan, Microsoft's chief scientist, said something similar. Metacognitive skills will matter most. Flexibility. Adaptability. Critical thinking. The ability to challenge assumptions and do deep thinking. She argued that a traditional liberal arts education is actually important for developing those skills.

Here is why I think both sides are right.

Karp is talking about who will thrive. Amodei and Teevan are discussing which skills will matter. And when you put them together, you get something very close to what I have been writing about for 338 editions.

The people who will thrive are the ones who think differently and bring uniquely human skills to the table. Pattern recognition. Strategic framing. Emotional intelligence. Creative problem-solving. The ability to see things from angles nobody else considered.

Those are neurodivergent strengths. And they are also humanity’s strengths. The Venn diagram overlaps a lot more than people realize.

Why This Validates What We Have Been Building

I have been writing this newsletter for over three years. Two thousand subscribers across fifty countries. Over 330 editions documenting how neurodivergent minds interact with AI differently.

The thesis has always been the same. Neurodivergent cognitive patterns are not a deficit requiring accommodation. They are a competitive advantage in AI collaboration.

In Edition 334, Anthropic's labor market study showed that the most exposed jobs involve sequential, repeatable cognitive tasks. The least automatable skills are exactly the ones neurodivergent thinkers tend to excel at.

In Edition 332, I introduced the Cognitive Balance Model. Human Initiation, AI Expansion, Human Integration. The framework for how humans and AI actually work together. These are the phases where neurodivergent strengths, especially lateral thinking and pattern recognition, are most valuable.

In Edition 323, I wrote about going against the grain my whole life. The loneliness of thinking differently. The choices nobody supported.

Alex Karp just told Fortune that thinking differently is the future. And a Gartner study says Fortune 500 companies are starting to recruit for it.

I am not saying I told you so. I am saying the world is catching up.

What This Means for Education

This connects directly to something I care about deeply.

Karp holds three degrees, including a JD from Stanford and a PhD in philosophy. And he is telling people those credentials are becoming less valuable in the AI economy. He said studying philosophy, his own field, is "going to be hard to market."

Meanwhile, Palantir is actively recruiting high school graduates who never went to college.

For homeschool families, this is significant. The traditional path of degree, credential, and career is no longer the only path. And for neurodivergent kids who have always struggled in traditional educational environments, that is liberating.

In Edition 337, I shared Jesse Genet's story of using OpenClaw agents to run her homeschool. In Edition 325, I told you about Makena using ChatGPT to prove me wrong about our cat's medication at age 14.

The Homeschool Parent's Guide to AI that I have been working on is built around exactly this idea. Your child does not need to fit into the traditional mold. AI gives them the tools to learn the way their brain actually works. And now a Fortune 500 CEO is telling the world that the way their brain works is exactly what the future demands.

What You Should Do With This Information

Here is the practical side.

If you are neurodivergent and working in a knowledge economy, this is your signal to lean in. Not to wait for someone to grant you permission. Not to keep hiding how your brain works. To actively position your cognitive difference as a professional strength.

If you are a parent of a neurodivergent child, share this article with them. A dyslexic billionaire is on record saying their way of thinking is the future. That matters for a kid who has been told their brain is broken.

If you are building AI skills, keep going. The Cognitive Balance Model from Edition 332 gives you the framework. The Cognitive Partner Membership from Edition 333 gives you the setup. The labor market data from Edition 334 tells you where the opportunities are. And now a Fortune headline is telling the world what you already know.

If you are an employer, pay attention. Gartner says 20% of Fortune 500 sales organizations will actively recruit neurodivergent talent by 2027. If your organization is not thinking about this, you are falling behind.

The Bottom Line

I have complicated feelings about Alex Karp. I imagine many of you do too.

But here is what I do not have complicated feelings about.

A dyslexic CEO told Fortune magazine that neurodivergent thinkers are one of only two groups who will succeed in the AI era. A major research firm says Fortune 500 companies are about to start recruiting for it. Anthropic's own data shows that the most automatable jobs are the sequential, repeatable ones, and the most resilient skills are those neurodivergent minds tend to be built for.

This is not wishful thinking anymore. This is the mainstream conversation.

And if you have been reading this newsletter, you have been preparing for it longer than most.

Have a great weekend!

• Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

Sources and Further Reading

Fortune: "Palantir's billionaire CEO says only two kinds of people will succeed in the AI era" (March 24, 2026) fortune.com/2026/03/24/palantir-ceo-alex-karp-two-people-successful-in-ai-era-vocational-skills-neurodivergence-gen-z-career-advice/

Gartner: 20% of Fortune 500 sales orgs will recruit neurodivergent talent by 2027 gartner.com/en/newsroom/press-releases/2024-02-29

Palantir Neurodivergent Fellowship jobs.lever.co/palantir/61eaa54c-e1b7-4064-afad-f7df3d48d652

Previously: Edition 337 covered OpenClaw and autonomous agents. Edition 336 covered GPT-5.4. Edition 335 covered cognitive partnership. Edition 334 covered Anthropic's labor market study. Edition 333 introduced the Cowork session, Import Memory, and the Cognitive Partner Membership. Edition 332 introduced the Cognitive Balance Model.

TL;DR- For My Fellow Skimmers

💬 The Quote: "There are basically two ways to know you have a future. One, you have some vocational training. Or two, you're neurodivergent." Alex Karp, CEO of Palantir, dyslexic, in Fortune magazine.

📊 The Data: Gartner predicts 20% of Fortune 500 sales orgs will actively recruit neurodivergent talent by 2027. Not as a diversity initiative. As a strategic advantage.

🏢 Palantir Is Hiring For It: They have a Neurodivergent Fellowship and a Meritocracy Fellowship for high school grads. They are putting money behind this.

🤝 The Counterpoint: Anthropic's Daniela Amodei says humanities skills, communication, emotional intelligence, curiosity, will be more important than ever. This complements Karp, not contradicts him.

⚖️ Both Are Right: The people who will thrive think differently AND bring human skills. That Venn diagram overlaps with neurodivergent strengths more than people realize.

📈 Validation: Anthropic's labor market study (Edition 334) shows the most exposed jobs are sequential and repeatable. The most resilient skills are lateral thinking, pattern recognition, and strategic framing. Sound familiar?

🎓 Education Impact: Karp holds three degrees and says they are becoming less valuable. Palantir recruits high school grads. For homeschool families, the traditional path is no longer the only path.

🧠 Your Move: Lean into your cognitive difference. Position it as a professional strength. Build AI skills. The Cognitive Balance Model (Edition 332) and the Cognitive Partner Membership (Edition 333) give you the tools.

🎯 Bottom Line: This is not wishful thinking anymore. This is the mainstream conversation. A Fortune headline is saying what you have known for a while. The world is catching up.

What You'll Learn Today

In this edition:

• What OpenClaw is and why the AI community has been obsessed with it

• How we set it up on an old Mac Mini and what we learned

• Why you can now talk to your AI agent through iMessage, Telegram, Slack, or Discord instead of opening Claude or ChatGPT

• The security considerations you need to understand before trying this yourself

• You do not need a Mac Mini. Phones, virtual servers, and cloud hosting all work too

• How a homeschool mom built a team of five AI agents to run her family's education

• Why this is the agent moment I have been waiting for and what it means for cognitive partnership

• A preview of the Homeschool Parent's Guide to AI coming soon

Reading Time: 15-18 minutes | Listening Time: 12-15 minutes if read aloud

I Have Not Been Talking About This. But I Should Have Been.

For the last month, in the AI circles I follow, there has been one topic dominating the conversation.

OpenClaw.

If you have not heard of it, that is okay. Most people outside of the developer and AI builder community have not. But it hit 247,000 GitHub stars in under four months. That kind of adoption does not happen unless something real is going on.

And something real is going on.

OpenClaw is an open-source platform that turns a regular computer into a 24/7 autonomous AI agent. You install it on a machine. It connects to your messaging apps. And then you just talk to it. Through Telegram. Through Discord. Through Slack. Through iMessage. Through whatever chat platform you already use.

You do not open Claude. You do not open ChatGPT. You do not open any AI interface at all.

You just text your computer. And it does the work.

What We Did

We uploaded an instance of OpenClaw onto an old Mac Mini. And we have been testing it.

I am going to be honest with you. There are a lot of little nuances to getting it set up. It is not plug and play yet. You need to be comfortable in the terminal or at least willing to learn. You need to create a separate user account on the machine with its own Apple ID. You need to configure the messaging bridges, install the skills you want, and set up the memory system.

It is getting easier with time. The setup wizards are improving. The documentation is better than it was even a few weeks ago. But this is still early.

The good news? Once it is running, the experience is something else entirely.

You give your agent an email address. You give it a chat. And you just talk to it. Through your phone. While you are at the grocery store. While you are coaching Little League. While you are doing anything else.

It works in the background. On your schedule. 24 hours a day if you want it to.

You Do Not Need a Mac Mini

I want to be clear about something. You can run this on any kind of machine. It does not have to be a Mac.

People are running OpenClaw on old laptops. On Linux boxes. Some people have even hooked it up to small language models running on a phone. If you are doing this for business or bigger operations, you are going to want a dedicated, more powerful machine. But for personal use and testing, almost anything works.

And for anybody who does not have a spare machine at all, you can set it up on a virtual server. That is a whole other technical conversation, but basically you can host this in the cloud the same way Google Docs or any other cloud service runs. You do not need your own hardware. And that option is becoming very popular because it removes the need to keep a physical machine running in your house.

The point is: the barrier to entry is lower than you think. And it is getting lower every week.

The Security Part You Need to Hear

I need to be direct about this.

This is not something I suggest you play with until you understand the risks.

And that is not just me being cautious. That is what every professional in this space is saying.

When you give an autonomous agent access to your file system, your email, your calendar, and your messaging apps, you are giving it a lot of power. If it is misconfigured, or if a prompt goes wrong, it can access things you did not intend.

Here is what the community recommends and what we are doing.

Use a dedicated machine or virtual server. Do not run this on your personal computer. An old Mac Mini, a cheap laptop, or a cloud instance. It becomes a server, not a personal device.

Create a separate user account. Separate Apple ID. Separate email. Separate Google account. Share only the specific files and documents the agent needs. Nothing personal.

No personal files on the machine. No photos. No financial documents. No social media apps. This is a work server.

Understand prompt injection risks. If your agent processes emails or web content, malicious instructions embedded in that content could potentially cause the agent to take actions you did not intend. This is a real and active area of security research.

Start small. Do not give it access to everything on day one. Start with one or two skills. Watch how it behaves. Expand from there.

The power is real. The risks are real. Respect both.

People Are Not Waiting for Business. They Are Starting at Home.

Here is something that surprised me about how people are actually using this.

A lot of people are not jumping straight into business applications. They are setting up OpenClaw for their home first. Running household tasks. Managing family schedules. Organizing files. Automating the little things that eat up time every day.

They are testing it in a low-stakes environment before they are willing to install it in their business. And honestly, that is smart. You learn how the agent behaves. You figure out the quirks. You build trust. And then when you are ready to use it for professional work, you already know what you are doing.

This is actually very similar to how I described learning AI in general back in Edition 331. Start where it is safe. Build expertise. Then bring it to the higher-stakes environment.

A Homeschool Mom Built a Team of Five Agents. This Is Worth Your Time.

I want to point you to something that blew my mind this month.

Jesse Genet, a former YC-backed startup founder and homeschool mom with four kids under five, went on The Cognitive Revolution podcast and described how she built a team of five OpenClaw agents. Each one runs on its own Mac Mini. Each one has a name, a role, and a personality.

Sylvie plans the homeschool curriculum. Claire acts as her AI chief of staff. Cole handles development and engineering projects. Theo creates content. Finn manages finances.

She talks to them through Slack. She sends voice notes from her phone while she is with the kids. They coordinate with each other, plan projects together, and even use her physical printer and 3D printer.

Here is the part that stopped me.

She photographed every educational toy she had ever bought for her kids. She had her agent build a complete inventory. Then she had Sylvie create a 70-lesson math curriculum for the year and insert the specific supplies she already owned into each lesson plan.

She had the agent analyze screen recordings of her kids' learning sessions and identify exactly where each child was struggling. The AI could tell that her son was confusing his sixes and nines. It tracked every math problem across every session.

She even had Sylvie create custom lesson plans for her mother, based on her mom's interests and hobbies, so grandma could teach the kids once a week without feeling overwhelmed.

And she was honest about the mistakes too. On day one, her agent Claire decided an email was urgent enough to draft and send a reply on Jesse's behalf, signed with her name. Without permission. The email was perfectly written. But it was a clear violation of trust. So Claire went to read-only mode for a while. Jesse treated it the same way she would treat an employee who overstepped. Guardrails first. Trust over time.

If you want to listen to the full conversation, it is on The Cognitive Revolution podcast. The episode is called "Try This at Home: Jesse Genet on OpenClaw Agents for Homeschool and How to Live Your Best AI Life." I highly recommend it.

cognitiverevolution.ai/try-this-at-home-jesse-genet-on-openclaw-agents-for-homeschool-how-to-live-your-best-ai-life/

Something I Have Been Working On

Jesse's story hit close to home for me. Because I have been working on something related.

The Homeschool Parent's Guide to AI.

It is a practical, non-technical resource for homeschool parents who want to help their kids learn, create, and think with artificial intelligence. Not a computer science textbook. A parent-friendly roadmap with copy-and-paste prompts, age-appropriate suggestions, and activities you can try today.

It covers choosing the right tools, using AI as a learning partner across every subject, teaching critical thinking about AI output, building an AI curriculum, and handling safety and ethics.

I wrote it because this is the guide I wished existed when Makena first asked me to teach her about AI. In Edition 325, I told you about the day she used ChatGPT to prove me wrong about our cat's medication. She was 14. She was dyslexic. She learned the answer from vet shows, not school. And AI helped her bridge the gap between knowing and showing.

That moment is what this guide is built around. The idea that AI is not replacing your child's learning. It is multiplying it. And that homeschool families have a massive advantage here because we have the flexibility to integrate new tools immediately without waiting for curriculum committees or school board approvals.

If You Liked Cowork, This Is the Next Level

In Edition 333, I told you about the four hours I spent in Cowork and how it changed everything. Cowork is Anthropic's version of this concept. It is a desktop agent that runs on your Mac, executes tasks autonomously, and you approve each step.

Cowork is the safer, more controlled version. Human-in-the-loop. Sandboxed. Built by Anthropic with their safety-first approach. If you are new to autonomous agents, Cowork is where you should start.

OpenClaw is the open-source version with more power and more risk. It connects to more platforms. It can run more independently. It has a skills system that lets you add capabilities like image generation, PDF processing, speech-to-text, and more.

And here is the exciting part. Anthropic just released Claude Code Channels, which brings OpenClaw-style behavior into Claude's official toolchain. You can now message Claude Code through Telegram and Discord. Persistent sessions. Always on. No dedicated Mac Mini required.

Claude Dispatch takes it even further. You send a mission from your phone. Claude executes it on your Mac. You come back and it is done.

The lines between these tools are blurring. And they are all converging on the same idea.

You should not have to open a chat window to talk to your AI.

It should just be there. In your messaging apps. In your workflow. Running in the background. Available when you need it.

Why This Is the Agent Moment

I have been writing about this concept for a long time. The idea that eventually you would not need 25 different AI tools. You would need one intelligent layer that handles everything behind the scenes.

In Edition 333, I described the problem. Twenty-five tools. Zero memory. Cognitive tax every single day.

In Edition 335, I talked about building your cognitive profile so AI knows how you think.

OpenClaw is the next step. It is the moment where the agent stops being a tool you visit and becomes infrastructure that runs for you.

You tell it what you need through a text message. It figures out which tools to use. It processes the files. It checks its memory. It runs the task. And when it is done, it texts you back.

When it is done right, it becomes self-learning. It gets better with every run of whatever you are doing. It starts remembering patterns. Becomes more recursive. More refined. In theory, it should just continue to get better with each pass.

That is the Cognitive Balance Model in its most evolved form. Human Initiation happens through a text message. AI Expansion happens autonomously in the background across whatever tools are needed. Human Integration happens when you review the output on your phone.

You are the director. The agent is the workforce. And the interface is just a conversation.

What This Means for Dyslexic Thinkers

Here is where I bring it back to us.

Every edition of this newsletter, I try to connect the latest AI developments back to why they matter for people who think differently. Sometimes that connection is a stretch. This time it is not.

This is the single interface moment.

For three years, the biggest source of cognitive friction in my workflow has been switching between tools. Opening Claude for one thing. ChatGPT for another. Gemini for research. Notion for notes. Beehiiv for the newsletter. Each one requiring me to re-orient, re-explain, and re-focus.

For a dyslexic brain, every context switch is expensive. Every new interface is a cognitive load event. Every time I have to figure out where something lives or how a tool works, that is brain budget I am not spending on the actual thinking.

An autonomous agent that lives in my messaging app and handles the tool-switching behind the scenes? That eliminates the friction at the source.

I do not have to go to the tools. The tools come to me. Through a conversation. The way my brain has always preferred to work.

In Edition 324, I wrote about how voice-to-text is not a convenience. It is an accessibility feature. Autonomous agents are the next version of that same idea. The interface disappearing so the thinking can happen.

The Practical Takeaway

If you are technical and curious, OpenClaw is worth exploring. The community is active. The documentation is improving. And the Mac Mini setup is becoming the standard approach, though virtual servers and cloud hosting are equally valid.

If you are not technical, do not worry. This is coming to you through easier channels. Cowork is already available. Claude Code Channels just launched for Telegram and Discord. Claude Dispatch lets you send missions to your Mac from your phone. The accessible versions of this are arriving fast.

Either way, here is what I want you to take away.

The era of opening a chat window to talk to AI is ending. The next phase is AI that runs in the background, reaches you through the apps you already use, and handles the complexity so you do not have to.

And for brains that have always found the interface to be the hardest part? That is the best news we have gotten yet.

This edition marks a shift.

I have heard you. The tool reviews, the research papers, the framework deep dives. Those matter and they will continue. But I also want to get back to being more practical. Helping you understand specifically why these tools are for you. How you can use them. What changes for your brain when you try them.

The workbench I have been envisioning. Where you put in your preferences and get a list of the best tools for how you think. Where the agent handles the tool selection behind the scenes. Where you just focus on the work.

That is not a future vision anymore. It is a Tuesday afternoon on a Mac Mini in my office. And it is a homeschool mom in Los Angeles running her family's education through Slack voice notes to five named agents.

We are living in it.

Previously: Edition 336 covered GPT-5.4 and the acceleration of AI. Edition 335 covered cognitive partnership and building your cognitive profile. Edition 334 covered Anthropic's labor market study. Edition 333 introduced the Cowork session, Import Memory, and the Cognitive Partner Membership. Edition 332 introduced the Cognitive Balance Model.

Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR- For My Fellow Skimmers

🤖 OpenClaw: Open-source platform that turns a computer into a 24/7 autonomous AI agent. 247,000 GitHub stars in four months. The AI community has been obsessed.

💻 Not Just Mac Mini: Works on any machine, virtual servers, cloud hosting, and even phones with small language models. The barrier is lower than you think.

💬 Text Your AI: Give it an email and a chat. Talk to it through iMessage, Telegram, Slack, Discord. No need to open Claude or ChatGPT.

🏠 Home First: Many people are testing agents for household tasks before deploying for business. Smart approach. Build trust in low-stakes environments first.

👩‍👧‍👦 Jesse Genet's Story: Homeschool mom, former YC founder, built five named OpenClaw agents to run her family's education. Inventory of every educational toy. Custom 70-lesson curriculum. AI analyzing her kids' learning sessions. Listen to her episode on The Cognitive Revolution podcast.

⚠️ Security First: Dedicated machine. Separate user account. No personal files. Understand prompt injection risks. Start small. This is powerful but not risk-free.

📚 Coming Soon: The Homeschool Parent's Guide to AI. Practical, non-technical, parent-friendly. Copy-and-paste prompts. Age-appropriate suggestions. Founding members get early access.

🔄 Cowork Connection: If you liked Cowork from Edition 333, OpenClaw is the more powerful, more open, more risky version. Claude Code Channels and Dispatch are bringing similar features into Anthropic's safer toolchain.

🧠 The Agent Moment: One interface. One conversation. AI handles the tool-switching behind the scenes. Gets better with every run. Self-learning. Recursive. The Cognitive Balance Model in its most evolved form.

♿ Dyslexic Advantage: Every context switch costs cognitive energy. An agent in your messaging app eliminates that friction at the source. The interface disappears so the thinking can happen. (See Edition 324.)

What You'll Learn Today

In this third of three articles covering the biggest week in AI this year:

• What GPT-5.4 actually is and why it matters

• What a million token context window means in plain language

• Why ChatGPT for Excel should get your attention about knowledge work

• How the pace of AI releases connects to the labor market data from Edition 334

• Why staying adaptable matters more than mastering any single tool

• How to think about all of this as a neurodivergent person building a cognitive partnership

Reading Time: 10-12 minutes | Listening Time: 8-10 minutes if read aloud

Where We Are

This is Part 3 of 3 from the biggest week in AI this year.

In Edition 333, I spent four hours in Cowork, mapped three years of my business, and told you about the Memory Tool and Import Memory. I launched the Cognitive Partner Membership. I counted 25 tools. None of them knew who I was.

In Edition 334, Anthropic's labor market study validated everything. The data showed which jobs are most exposed, why the window to position yourself is closing, and why cognitive flexibility is your real job security.

In Edition 335, I went deeper on what cognitive partnership actually looks like now that AI can remember how you think. Your brain has a profile. Build it. Protect it. Carry it with you.

Today is about speed. About acceleration. About what happens when every few weeks the tools get meaningfully better and what that means for you.

The Release

On March 5th. the same day Anthropic published their labor market study. OpenAI released GPT-5.4.

It comes in three versions. A standard model. A reasoning model called GPT-5.4 Thinking for complex professional tasks. And a high-performance version called GPT-5.4 Pro for the most demanding work.

If you are not a developer, here is what actually matters about this release.

A Million Token Context Window

The headline feature is a one million token context window.

In plain language, that means GPT-5.4 can process an enormous amount of information in a single conversation. We are talking about entire document libraries. Hundreds of pages of reports. Complete codebases. Full project histories.

To put it in perspective, that is roughly the equivalent of reading several thick novels in one sitting and being able to discuss any detail from any of them.

Why this matters for you.

If you have ever hit the wall in an AI conversation where it says it cannot process any more, that wall just got pushed way back. You can now feed it your entire project, all your notes, your research, your context, and have a meaningful conversation about all of it at once.

Think about what I described in Edition 333. That four-hour Cowork session where I mapped three years of work. A million token context window means that kind of deep, comprehensive session becomes possible for even larger, more complex projects. You can bring everything to the table and the AI can hold all of it in its head while you talk through it.

For neurodivergent thinkers who struggle with organizing and synthesizing large amounts of information, this is a significant upgrade. Our brains are great at seeing patterns across large datasets. But the sequential organization of that data? That is where the friction lives. A million token context window lets AI handle the organization while we do the pattern recognition.

ChatGPT for Excel

This one is easy to overlook but I think it is one of the most important things that happened this week.

OpenAI launched ChatGPT for Excel.

Think about what that means.

Excel is the backbone of knowledge work. Budgets. Reports. Data analysis. Project tracking. Financial modeling. If you work in an office, you probably touch a spreadsheet at least once a week if not every day.

Now AI is inside that tool.

Not as a separate app you switch to. Not as a chatbot in another tab. Inside the spreadsheet itself.

This is exactly what Anthropic's labor market study from Edition 335 was measuring. The gap between what AI can theoretically do and what people are actually using it for. Features like ChatGPT for Excel close that gap. They take AI from "something I use sometimes" to "something embedded in how I work every day."

And when that happens at scale, the labor market shifts that Anthropic is tracking will accelerate.

This also connects directly to the "meet in the middle" concept from Edition 331. The experienced professional who has been using Excel for 20 years and now has AI built into that tool? That person has a massive advantage if they learn to use it. The young worker who is AI-native but has never built a complex financial model? They need that domain expertise to make the AI actually useful.

The tool does not replace the knowledge. It amplifies it.

And here is the thing about Excel specifically. Anthropic also launched Claude in Excel. So now you have both major AI platforms integrating directly into the same spreadsheet tool. The competition between these platforms is making the tools better, faster, and more integrated. That pace benefits you.

The Pace of Releases

Let me zoom out for a second.

GPT-5 launched earlier this year. Since then we have seen GPT-5.1. GPT-5.2 with its Codex model for coding. GPT-5.3 Instant for faster everyday work. And now GPT-5.4 with the million token context window and Excel integration.

That is five major releases in the GPT-5 family in a matter of months.

On the Anthropic side, Claude Opus 4.6 launched in February with its own million token context window. Claude Sonnet 4.6 followed. Cowork launched for knowledge workers. Claude in Chrome, Excel, and PowerPoint. The Memory Tool and Import Memory that I covered in Edition 333. And the labor market study from Edition 334.

The pace is not slowing down. It is accelerating.

Every few weeks, the tools get meaningfully better. More capable. More integrated into the software you already use. More embedded in how work actually gets done.

In Edition 330, I asked what problem solvers do when AI solves all the problems. The pace of these releases is why that question is not hypothetical anymore. We are watching AI get closer to solving more problems in real time.

What This Means for Knowledge Workers

Here is the honest truth.

If your job involves processing documents, analyzing data, writing reports, managing spreadsheets, coding, customer service, or any combination of these things, the tools that can do those tasks just got significantly better.

That does not mean you are getting fired tomorrow. The Anthropic study from Edition 335 shows that unemployment has not increased for exposed workers.

But it does mean the bar is rising.

The baseline expectation for what one person can produce is going up. If AI can help you analyze a dataset in 10 minutes that used to take 3 hours, your employer is going to start expecting that speed. If AI can draft a report that used to take a full day in 30 minutes, the definition of "productive" changes.

This is not about replacement. It is about amplification.

The people who learn to use these tools will produce more. Get more done. Have more capacity. And the people who do not will look slower by comparison.

That is why the message from this entire series matters so much. Edition 333 gave you the setup. Edition 334 gave you the data. Edition 335 gave you the cognitive partnership framework. And now Edition 336 is showing you the pace. The window that Anthropic's data showed is still open? Every release like GPT-5.4 makes it a little smaller. Not because people are getting laid off. Because the standard for what "good work" looks like is quietly going up.

How to Think About This as a Neurodivergent Person

Here is what I keep coming back to.

Every one of these upgrades makes the AI better at the sequential, structured, detail-oriented tasks that create the most cognitive friction for neurodivergent minds.

Better at processing documents. Better at organizing data. Better at formatting. Better at following step-by-step procedures. Better at keeping track of complex details across long conversations.

Those are the exact things that drain our energy the fastest.

And every time AI gets better at those things, it frees up more of our cognitive energy for the things we are actually good at. The pattern recognition. The strategic thinking. The creative connections. The big picture stuff that AI still cannot touch.

The Cognitive Balance Model gets more powerful with every release. Because the AI Expansion phase keeps getting stronger. Which means Human Initiation and Human Integration become even more valuable.

And with the cognitive profile work from Edition 335, that AI Expansion is now personalized to how you specifically think. It is not just a better tool. It is a better tool that knows you.

Your job is not to compete with AI on the tasks it is getting better at.

Your job is to get better at the tasks AI cannot do.

And if you are dyslexic, if you are neurodivergent, if your brain has always worked differently, you have a head start on exactly those tasks.

The Practical Takeaway

Do not try to learn every new model the week it comes out.

That is a recipe for burnout and it is not the point.

Instead, focus on this.

Pick one AI tool and get really good at using it for your actual work. Not for playing around. For the real tasks that take up your day. Whether that is Claude, ChatGPT, or something else, go deep with one.

Pay attention to where AI shows up inside tools you already use. Excel. Chrome. Slack. PowerPoint. The integration into existing workflows is where the real transformation happens. Not in the standalone chatbot.

Keep your skills portable. This is why Edition 334 matters. Build your cognitive profile. Use Import Memory. The specific tool will change. Your ability to work with AI effectively is what carries forward.

Stay curious, not anxious. Every upgrade is an opportunity to offload more cognitive friction. That is a good thing. Especially for brains like ours.

If you want the guided path, the Cognitive Partner Membership from Edition 334 is still open. Fifty founding member spots at $19/month locked forever. The 30-day onboarding walks you through all of this step by step so you do not have to figure it out alone.

The Bottom Line

GPT-5.4 is impressive. The million token context window is a real capability jump. ChatGPT for Excel is the kind of integration that changes daily workflows for millions of people.

But here is what I want you to take away from this entire three-part series.

Edition 333 gave you the tools and the membership. Import Memory. The Memory Tool. Cowork. The Cognitive Partner Membership.

Edition 334 gave you the data. The Anthropic labor market study. The gap between capability and adoption. The generational crossroads.

Edition 335 gave you the framework. Your cognitive profile. How the Cognitive Balance Model gets stronger when AI knows you.

Edition 336 showed you the pace. GPT-5.4. Five major releases in months. ChatGPT for Excel. The acceleration is real.

Together, they tell you one thing: the time to act is now.

Not next year. Not when it feels less overwhelming. Not when someone makes you. Now.

The acceleration is real. The opportunity is real. And if you have spent your whole life thinking differently, adapting to systems that were not built for you, and finding creative workarounds for everything...

You have been preparing for this moment your entire life.

This is Part 3 of 3 from the biggest week in AI this year.

Previously: Edition 335 covered cognitive partnership and your cognitive profile. Edition 334 covered Anthropic's labor market study. Edition 333 introduced the Cowork session, Import Memory, the Memory Tool, and the Cognitive Partner Membership. Edition 331 explored the generational crossroads. Edition 330 asked what problem solvers do when AI solves everything.

Coming soon: More on how to put all of this into practice. The frameworks, the tools, and the strategies for staying ahead. If you are a founding member, the 30-day onboarding has already started.

• Matt "Coach" Ivey Founder, LM Lab AI | Creator, The Dyslexic AI Newsletter

Dictated, not typed. Obviously.

TL;DR (Too Long, Didn't Read)

🚀 GPT-5.4: Three versions. Standard, Thinking, and Pro. Dropped the same day as Anthropic's labor market study.

📚 Million Tokens: Can process entire document libraries in one conversation. Think of it as the Edition 333 Cowork session but for even bigger projects. Huge for neurodivergent thinkers who see patterns but struggle with sequential organization.

📊 ChatGPT for Excel: AI is inside the spreadsheet now. This closes the gap Anthropic measured in Edition 333 between capability and actual usage. Claude in Excel exists too. Competition makes the tools better for everyone.

⚡ The Pace: Five major GPT-5 family releases in months. Anthropic matching that pace. The question from Edition 330 about AI solving everything gets less hypothetical by the week.

📈 Rising Bar: Productivity expectations are going up. Not replacement. Amplification. The "meet in the middle" from Edition 331 is more urgent than ever.

🧠 Neurodivergent Advantage: Every upgrade makes AI better at the sequential tasks that drain our energy. That frees us up for the strategic, creative, big-picture work AI cannot do. The Cognitive Balance Model from Edition 332 gets stronger with every release.

🔑 Cognitive Profile Matters: The pace of new tools is why Edition 334's message about building your cognitive profile is so important. Tools change. Your portable AI partnership carries forward.

🚀 Founding Members: The Cognitive Partner Membership from Edition 334 walks you through all of this. 50 spots at $19/month locked forever. 30-day onboarding has started.

🎯 Bottom Line: The data says the window is open. The tools show how fast the landscape is moving. The time to act is now. If you have spent your life thinking differently, you have been preparing for this moment.