Good morning.

I've been following a stay-at-home mom on Instagram since July.

She's been taking content seriously for a few months, filming on her phone, and one post changed everything for her.

Watching her content, it finally clicked why what I've been doing intentionally has actually been holding me back.

Today, let's talk about:

• The 7 words that kicked off her biggest post

• The part of my process I shouldn’t have handed to AI

• A simple, 3-step framework you can run this week

Let’s get into it.

— Gordon

📱 Why I Followed a Stay-at-Home Mom on Instagram

I wouldn’t consider myself the target audience for a self-proclaimed “non-techy, Jesus-lovin’, stay-at-home mom of 3.”

Instagram posts about Costco hauls, make-up routines, and slow mornings before the kids wake up aren’t a match for my algorithm.

Frankly, I’m not the slightest bit interested.

But, here’s the thing.

The Instagram algorithm is laser-focused (pew pew) on finding the perfect user-to-content match.

It does that by analyzing the posts you engage with (e.g., share, save, comment, like, and view) and by tracking your watch time on Reels.

The more you engage and the longer you watch, the more Al G. Rhythm believes you’re interested in those specific topics.

When new content about those topics is posted, it’s instantly added to the feeds of users who are the best match for that content.

That’s exactly how @the.jessicacarter’s “How I Edit My Yapping Reels” post series hit my feed.

Since early May, when I was writing a short newsletter series about my journey to create and post my first Instagram Reel, I’ve been wanting to learn how to edit with Instagram’s Edits app.

I know how to edit in Adobe Premiere Pro on a computer, but every app is different, and learning to edit in a new app and on a phone felt really overwhelming for me.

Even as a “tech guy,” learning new apps is like learning a new language dialect. There are similarities, but it’s definitely not the same. It takes time.

Jessica’s short series broke down her editing process in the simplest, easiest-to-understand way.

It felt just like sitting down with a trusted friend and having them walk you through something in a comfortable, approachable way that didn’t make you feel bad about yourself, unlike most other posts that teach.

I immediately had an aha moment and understood the simple workflow I had been dreading learning (again, even as a techy).

That’s what caught my attention.

She was showing me how to create tech content the same way I already do with people in person.

I just hadn’t been able to visualize how that looks online, because few creators, if any, teach like that.

After watching her 3-Reel editing series (now 5 parts), I watched a few other videos and followed her account.

Instagram post

Her content over the next week was transparent, honest, self-deprecating, and felt really genuine. (The post above is exactly what filming really looks like, btw.) 😂

That’s the stuff I’m interested in that the algo can’t see.

The series she didn’t plan

What the algorithm also couldn’t see is that she never planned to make a series of Reels about video editing.

Part 1 opens with 7 words.

“Alright, you asked, so here we go.”

People kept asking how she edited her Reels, so she answered.

Her entire setup is her iPhone’s front camera, the free Edits app and its teleprompter, a $79 DJI Mic Mini, a cheap ring light, and 2 fonts.

The Edits app is the same one I’d been too overwhelmed to learn.

Part 1 did 311,000 views. That’s roughly 20x her average post.

A few weeks later, she posted her July income recap. It got 12,000 views.

Her post about the money got 4% of the views her post answering a question did.

She has one digital product, a $9 Trello template she made because people kept asking how she organized her content. It’s the same template she actually uses herself.

It sold 541 times and made up most of her $6,683 month.

Without a planned launch. 🤯

Jessica isn’t just creating content for her own gain.

She’s building a community. She’s paying attention to what resonates.

The idea that changed everything was sitting in her comments.

What I built instead

Ace is on the case 🐬

For years, I’ve been trying to crack the code on producing high-performing content in the least amount of time.

I know how many hours it actually takes, and business owners don’t have that kind of time.

Solving that would be real leverage, and something I could hand them.

That was the thing I cared about most.

So I chased the shortcut. Constantly.

In 2018, I started filming on my iPhone.

Then I bought a Sony A7iii, learned Premiere Pro, and kept buying gear until my office had turned into a studio. Roughly $12,000 of it, all to chase a look I loved on someone else’s channel.

Congratulations, I’m an idiot. 🥳

Then I did the same thing to my writing.

I took the proven script structures I’d learned in Kallaway’s Short-Form Academy and used AI to fit my story into them.

The structures work. That’s why they’re proven.

They were also far too much for someone who had never published a single Reel.

2 scripts died in June. I wrote that whole newsletter series about making that first Reel, then killed both scripts after filming.

Still haven’t posted one. FML. 🫠

I handed AI the one part that had to be mine.

I had the expertise I’d earned sitting right there. Then I let AI write it in someone else’s style.

Teaching from inside the learning

The creators I learn from break down their process, but they teach it from where they sit now, after it's already worked for them.

Their step 1 assumes the camera stopped being scary years ago.

For someone starting, being on camera is the hurdle, and every single thing stacked on top of it is one more reason to never start.

Jessica taught from inside the learning instead.

Trial and error, in real time, and still figuring it out.

She was doing for editing exactly what I do for technology.

Translating it for people who aren’t technical, in plain words, without making them feel behind.

That’s why it clicked. I recognized the thing I do in person, and it was working for her on camera. 💡

I still haven’t cracked the code on making great content fast, by the way.

Advice built for year 3 is a massive wall on day one.

The algorithm showed me her account because of the editing videos I watch.

It had no idea why I stuck around.

That’s the stuff I’m interested in that the algo can’t see.

Here’s what I learned from Jessica, and you can run with it this week.

My FAQ Framework:

1. Find the question people keep asking you

2. Answer it plainly, for free, in public

3. Offer a real shortcut for cheap

Content and entrepreneurship aren’t one-size-fits-all.

You deserve to build it in a way that truly fits you.

What’s the question you’ve answered 20 times and never made the thing for?

Hit reply and share with me.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

🎬 Instagram shares tips on video creation
Instagram asked its top creator consultants what they actually do differently, and the answer is a 5-step workflow you can run in Edits.

📺 The new fundamentals of brand strategy for modern social media
A 20-year-old social strategist argues that the individual post stopped being the unit that matters and names what replaced it.

🌡️ 6 Ways Leaders Harness Stress
Executive stress is running higher than it did at the peak of the pandemic. There are 6 default reactions to it, and knowing yours changes how you decide under pressure.

Good morning.

This week, I returned to my original newsletter format, just for fun.

Feels a little nostalgic. 🤓

10 months ago, I typed 3 complaints into a chat window at 8 in the morning. I had no idea I was starting a list.

That list is now the first thing my AI reads, and it's longer than the instructions I give it.

Today, let's talk about:

• Why writing longer prompts makes your AI output worse

• The ChatGPT fix that quietly created my next 2 rules

• The 5 steps to build a list that makes AI sound like you

Let’s get into it.

— Gordon

🚫 What to Tell AI It Can’t Do

On September 9th, 2025, at 8:24 in the morning, I was 49 days into a Twitter thread streak that would run 210 days.

I stopped mid-draft and typed 3 complaints to ChatGPT (I called it “Chet”).

“No rhetorical questions. Stop saying ‘compounding.’ That post is weak.”

For 49 days, I’d been fixing the same kinds of lines by hand.

My answer every time was to add more instructions. A longer prompt, more context, and another reminder of what I did or didn’t want.

The chat just got longer. The quality never got closer.

I’d end up rewriting everything myself, which meant I did the work twice.

The dogshit draft was on my calendar. The rewrite wasn’t.

It cost me 30 minutes I hadn’t scheduled, every single time. 🫠

The fastest way to sound like yourself is to write down what you’ll never say.

Every prompting guide teaches addition.

Give it a role, give it context, give it examples, describe your tone.

Act like an expert in my field. Here’s my business. Here are 3 posts I like. Keep it warm.

Instructions tell the AI what to do. They rarely tell it what it shouldn’t do, so it fills the gaps with the average of everything it’s ever read on the internet.

That average is where rhetorical questions live.

The rules that actually fixed my output were subtractions.

I’ve been collecting them for 10 months, and the list of things I never say is now longer than the list of things I ask for.

When you’re always short on time, you can’t afford to do the same work twice.

HERE’S HOW:

1. Pull 3 Things You Wrote That Sound Like You

2. Call Out What Makes Them Good

3. Turn Your Rejects into Nevers

4. Add the Topics You’ll Never Write About

5. Put the List Where Every Session Starts

Steps 1 and 2 build the thing your “nevers” protect.

Skip them, and you’ll write a list of rules that makes your writing quieter instead of sharper.

Step 1: Pull 3 Things You Wrote That Sound Like You

Pick the 3 that sound the most like you talking.

Use these filters.

• Published over drafted, because you already stood behind it

• 3 different formats, like an email, a social post, and something long-form

• Written before AI instead of written with AI, so you’re reading yourself instead of a collaboration

Your voice is already sitting on the page. You’ve just never read it as data.

Step 2: Call Out What Makes Them Good

The positive half comes first for a reason.

A never only works when it’s protecting something specific.

Read your 3 samples and write down what you actually do.

Mine looks like this:

• Short fragments to close a beat

• A compressed line at the end of every section

• Parenthetical asides that undercut my own sentence

• Italics where the stress would land if I said it out loud

• One idea per paragraph, never more than 2 lines

Point at a real example every time.

If you can’t find one in your own writing, it’s a preference, not a rule.

Step 3: Turn Your Rejects into Nevers

Now, raid your rejects.

You don’t have to invent this list, because you’ve been building it every time you deleted a line without thinking about it.

My 3 complaints to Chet were 3 different kinds of never, and I didn’t notice that until I read them back 10 months later.

• A form you never use. No rhetorical questions.

• A standard every line has to clear. No filler posts.

• A word your reader can’t picture. Stop saying compounding.

Most prompting advice treats a ‘never’ like a word swap. It’s 3 different jobs.

Chet wrote my rules back to me as a list with a red ❌ next to each one, and that was the first time my never list existed anywhere outside my own head.

Ugh. My prompting was awful back then. Started from the bottom, for realz

Then it followed all 3 and broke 2 rules that didn’t exist yet.

The post I complained about started here.

Chet’s fix landed here.

The rhetorical question is gone. In its place, a second “Most owners” in a thread that already used it 4 posts earlier, and an em dash, the long dash, sitting where 2 short sentences used to be.

Both of those are banned in my rules today.

Reading the rest of that thread now, I catch 2 more.

“Most owners wait until the work slows down” is a “Most…” opening line, which I no longer allow. “Here’s the truth:” is a colon I’d now rewrite as a sentence. Same story.

6 rules, 1 thread, and I only saw 3 of them at the time.

Take something you wrote before you had rules and mark it up with what you know now.

One more thing about that rhetorical-question rule, because the first version was too blunt. For months, it was 3 words, and that didn’t work in every scenario.

The version I use now says where. Never in teaching. Fine in the closing question, where I actually want you to answer.

Every fix points at the next rule you need.

Step 4: Add the Topics You’ll Never Write About

The second half of your list has nothing to do with wording.

It’s about subject matter, and it’s the half that ends the blank page problem.

Write down what’s off the table for you.

• Frameworks you borrowed from someone else without crediting them

• Subjects adjacent to yours that aren’t actually yours

• Anything you’d have to fake authority to cover

• Claims you can’t prove

Mine has 6 lines on it. This morning, I killed an idea I liked because it sat next to my lane instead of in it.

Most ideas now get eliminated before I ever open a document, which sounds limiting right up until you try it. 🤌

A shorter list of things you’re allowed to say is easier to start from than an open field.

Deciding what you won’t cover is what makes it uniquely yours.

Step 5: Put the List Where Every Session Starts

A rule that lives in a chat dies with the chat. Mine went through 3 homes.

• A chat window: 3 complaints, typed once, gone the moment I started a new chat

• A playbook: One document I updated the same day a new problem showed up

• A project context file: Read before a word gets written, every session

That day-49 playbook had 11 rules. 6 of them were things not to do.

I didn’t plan that. I just started writing down what I’d rejected over and over.

Put yours wherever your AI reads instructions before you type.

GenAI tools all have a settings box, usually called custom instructions or project instructions. Most support project files too (that’s where mine are).

A document you paste at the top of every session with your initial prompt works too.

The location matters less than never having to remember it every time.

One of the main reasons I switched to Claude in December was the ability to create projects with specific instructions and context files for repetitive tasks. Chet [ChatGPT] now offers the same features.

Your list doesn’t have to be finished to start working.

Add the next thing you reject, and the next, and it gets sharper as you go.

I still rewrite lines. I just don’t rewrite the same ones every week.

Open the last thing AI wrote with you and find the one line you deleted without thinking about it.

You already know which one it is.

That’s your first never.

Hit reply and share with me what it was.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

🔬 Research: How AI Agents Broaden the Scope of Knowledge Work
Perplexity's own researchers expected faster work. They found people attempting work they'd never have taken on before.

📸 How to maximize Instagram marketing in 2026
Most of the work happens before you post. A research routine for finding what's already working in your niche.

🔓 PSA: Your Claude shared chats and Artifacts may have ended up on Google
If you've ever hit share on an AI chat, this one's worth 2 minutes in your settings this week.

Good morning.

Almost everyone has AI now. Far fewer have it running underneath the whole business, and that's where the results really come to life.

This week, I get into why, with the workflow I just rebuilt as the case study.

Today, let's talk about:

• What the 14% do that everyone else skips

• The AI editorial team I built, and how it's checking this very issue

• An AI photo critic that coaches your next shot

Let’s get into it.

— Gordon

🧩 Why 14% of Small Businesses Make AI Work

On Monday, I sat down with coffee and a light breakfast, reviewed my calendar and my weekly to-dos, checked my email for anything urgent, and opened Claude to write this newsletter.

I wasn't sure what it would be about yet.

A week earlier, I'd rebuilt my entire workflow from the ground up. I trusted it to guide me step by step, and began my 9:30 time block to get to work.

A blank page on a Monday used to be where I stalled out.

This time, I wasn't worried. That feeling was new.

The Way It Used to Work

For a long time, AI for me was a pile of one-off chats.

Back in the day, when I used ChatGPT, I wasn’t using custom instructions (I didn’t know how), there weren’t files to reference, and I had no idea how to write a model-optimized prompt.

The only thing that carried from one session to the next was the newsletter structure I kept in my head.

So every chat came out completely unique and very inconsistent.

The quality leaned on ChatGPT's memory (which was spotty at the time), so I had to reteach it how to do the work and how I wanted things done, over and over.

That repeated instruction caused my process to take much longer than it should have.

Worse, I rarely trusted the final newsletter I'd assembled at the end of it.

Reteaching the tool at the start of every session turned out to be slower than just writing from scratch.

Access Isn't the Problem

In 2026, Goldman Sachs found that 76% of small businesses now use AI.

However, only 14% have it built into their operations.

Almost everyone has access. Almost no one has integration.

Access is opening the app. Integration is the app already knowing your business before you type a word.

The 14% of businesses win for one reason, and contrary to last week's newsletter, it has nothing to do with running the right model. They stopped bolting AI onto single tasks (reaching for it on one job, then setting it back down) and wired it into how the work runs.

There's an MIT Sloan study that backs this up. AI raised the strongest operators' profits by about 15% and cut the weakest operators' profits by about 10% because the strugglers took generic AI advice with no system to filter it.

Access hands you a tool. Integration turns it into leverage.

What Changed on My End

Now Claude sits at the center of my day.

Each morning, I review my calendar and my task list, check my sales pipeline, then my email, before I drop into deep work in Claude, where my Projects are already set up and ready to rock.

Onward Everyday integrated context files

Two things make that possible, and they're easy to blur together.

My Project files hold the context, so Claude keeps a working memory of my business from one session to the next.

Separately, Connectors link Claude to my other apps so it can access tools like beehiiv and Notion directly.

The result is that I spend most of my day in one place instead of bouncing between a stack of separate apps.

Building that system cut my production time from 20 hours a week to 4-6 hours. Then learning to lead with my own thinking, instead of rushing to hand everything to AI, brought it down again to 3 or 4.

Those hours matter, but the biggest change was my confidence in the quality of what I ship each week.

That came from a 4-part editorial workflow I built.

Every issue now runs through a quality check, a senior editor pass, a copyedit, and a cold read.

It's like handing one draft to a managing editor, a senior editor, a copyeditor, and a fresh set of eyes. A team of specialists that a solo creator doesn't have.

The newsletter is still only as good as my ideas and my writing.

But I know those get scrutinized the way a larger publication would, and that's why I trust the newsletter that goes out each week.

Same tool everyone has. The process is mine, and that's the difference.

Building One That's Actually Yours

Here's how I recommend you approach this.

Build your own system instead of copying mine, in 3 steps:

1. Provide deep context about your business

2. Have it study how the leaders in your field operate

3. Shape the process around the way you already work

Generic in, generic out. The context is what makes it uniquely yours.

Mine started as a single quality check. I wanted more, but I wasn't sure what to add.

So I asked Claude how major publications run their editorial process, and together we built the senior editor pass and the copyedit.

A week ago, I added the cold read after being concerned that readers might get confused partway through a piece. That idea came from Kallaway's Short-Form Academy, and Claude helped me build the task.

That's the line between off-the-shelf AI and integrated AI.

One gives you an average answer.

The other gets shaped, again and again, until it fits only you.

This issue is the proof.

I sat down on Monday, not knowing what to write, and the system carried me here.

It's moving through those 4 editorial passes right now, before it reaches you.

Quality Check one, two, what is this

That's what integrated actually feels like. Not another underutilized tool, but a business that keeps moving you forward, even when you’re unsure where to start.

So here's my question for you this week: What's the thing you re-explain to your AI every single time you open it?

That's where your system wants to start.

See you next week.

ONWARD TOGETHER.

Next Steps 👇

PS — Whenever you're ready, here's how I can help

A free 30-minute discovery call. We'll pinpoint the 1-2 places in your unique business where AI would save you the most time, so you leave with a clear starting point. If you want help building it out, I'll show you what that looks like.

Everyday Insights 👀
A few things worth reading this week

🛠️ Frontline Workers Know How to Solve Your Organization's Biggest Problems
The people closest to the work spot are the first to spot what's broken. There's a disciplined way to surface those fixes, the kind standard surveys miss.

💧 AI has a water problem — Google thinks it has a fix
AI's compute runs on data centers, and those quietly run on water. Google now says it'll replenish more than it uses by 2030.

📸 Adobe Camera App's New Feature Will Critique Your Photos Using AI
Adobe's experimental camera app now plays photo critic, weighing your framing, lighting, and emotional impact, then coaching your reshoot. Handy if you shoot your own content.

Good morning.

For months, I did what every AI creator told me to do. Reach for the most powerful AI model every time.

It felt smart, and the output was better, but it was ultimately the wrong call.

Today, let’s talk about:

• Why reaching for the most powerful AI model is quietly costing you time and money

• A dead-simple way to know which model to use for what, in plain words

• The $1.5 billion bet that says knowing how to use AI beats having the best AI

Let’s get into it.

— Gordon

🎛️ Why the Best AI Model Isn't Always the Right One

We were at the dining table at a cabin when my brother-in-law told me his company had recently capped how much AI employees could use.

Naturally, the people who lean on it all day are furious. The people who barely touch it didn't even notice.

He's a product owner and IT manager, so he sees both sides of this every week.

He sees the real gap that large organizations are facing right now.

Everyone has access to a powerful tool. Nobody was handed a map for how to use it in their individual roles.

So, people aim it at whatever's in front of them. One CTO caught his team using an expensive model to check the weather.

Multiply that across an entire company, and you get the headlines. MIT found that 95% of AI pilots have zero measurable impact on the bottom line.

The tool works, but no one knows how to use it to move the business forward.

That gap is too big to fix in one email. So here's the most fixable corner of it: which model to use for what.

A guide I wished existed

Days before that conversation, I'd been thinking someone should make a dead-simple chart.

Which AI model to use, for which kind of job, in plain words.

Then Anthropic, the company behind Claude, published almost exactly that.

Almost.

Helpful. Sort of.

It's genuinely useful. It's also still written for people who already know what "long-horizon tasks" and "methodology critique" mean. Surely not I, Sir Robin.

The guy checking the weather doesn't need that. He needs the Monday-morning version, in plain words.

The mistake I made

For a long time, I ran Sonnet for almost everything, and it was great.

Then, every AI creator I followed started saying the same thing. Use the most powerful model, always. So I switched to Opus for everything. The results felt sharper, and I stayed there for months.

The catch is that even Anthropic's own guide tells you to start with Sonnet. The loudest advice online was pointing me away from the model built for most everyday work.

Better output felt like proof. It was really just a habit forming.

What changed

Recently, I started matching the model to the task rather than defaulting to the most powerful one every time.

What made me comfortable experimenting was one small change. You can now switch models inside a single chat, which wasn't possible before.

When I built a beehiiv landing page and a 31-email automation for another business, I used Opus to plan and build the hard parts. Then, when it came time to crank out all 31 emails, I switched to Sonnet and knocked them out faster, cheaper, and without burning through my usage.

Same project, 2 models. One for the thinking, one for the doing.

Reach for power when you're thinking. Reach for speed when you're doing.

The version I'd hand a friend

Strip the jargon, and it's 4 tools for 4 jobs. Here's how I actually use them:

1. Haiku is the fastest one. Fast lookups, summarizing a long thread, pulling numbers out of a document. It sips your usage. (Transparently, I've never really used it. For that stuff, I still reach for Google or just ask Sonnet.)

2. Sonnet is the daily driver, and where you should start. Writing, everyday research, drafting emails and content, running a plan you've already thought through. Most of your day lives here.

3. Opus is the deep thinker. Strategy, positioning, hard analysis, planning a build. It's my default for real work, but it leans harder on your usage, so I’d save it for your most important work (e.g., client work) or when the thinking actually needs it.

4. Fable is the most autonomous. You describe the outcome, and it plans and builds while checking its own work. So far, I've used it to rebuild this entire newsletter system end-to-end.

One thing worth knowing. On the free plan, you get Sonnet and Haiku. Opus and Fable start on the paid plans. So if you can't find them, that's why.

The trap isn't using AI too much. It's using the best model for work the fastest one would nail.

The limits my brother-in-law's company set are just the enterprise version of a challenge you already have. Every plan has a ceiling, and the biggest models reach it the fastest.

Matching the model to the moment is how you reduce the times you hit that ceiling for no reason.

This week, before your next session, ask one question. Am I thinking, or doing?

Then pick the model to match.

See you next week.

ONWARD TOGETHER.

Next Steps 👇

PS — Whenever you're ready, here's how I can help

A free 30-minute discovery call. We'll pinpoint the 1-2 places in your unique business where AI would save you the most time, so you leave with a clear starting point. If you want help building it out, I'll show you what that looks like.

Everyday Insights 👀
A few things worth reading this week

🛠️ Anthropic, Blackstone Bet the Next Trillion-Dollar AI Business Is Implementation, Not Just Models
Anthropic and some of Wall Street's biggest names just put $1.5 billion behind a bet that the real money isn't in better models. It's in helping companies actually use them.

🫠 AI Adoption Is Overloading Your Middle Managers
Leadership gets the strategy wins, the front line gets the productivity boost, and the people in between get handed everything nobody else wants to do.

💼 LinkedIn Shares Tips on Effective Posting
LinkedIn published its own checklist for what to post, how often, and how to get picked up by AI search. Straight from the source.

Good morning.

Last week, my AI output jumped in quality, and it had nothing to do with a newer model.

Same Claude. Better outputs.

Learning why has shifted how I work with AI.

Today, I'll walk you through what happened and share how to do the same in the app you already use.

We’ll talk about:

• Why the same model can feel sharper in one app than another

• The shift that stops your AI from getting worse the longer you work

• How to get better, steadier output in the app you already use

Let’s get into it.

— Gordon

🗄️ Why I Changed How I Use AI

I’m fundamentally changing how I use AI.

It happened by accident.

Last week, I moved one file, a set of instructions I’d written for Claude, out of the web app and into Claude Code on my desktop.

Then I turned a small team of AI agents loose on a research project (the accidental business plan from last week, if you caught it).

The very first response came back sharper.

A lot sharper.

More specific, more detailed, and more aware of nuance than anything I’d been getting in the web app.

I’ve used Claude nearly every day since December. I know its rhythms.

This was different enough that I stopped and paid attention.

The tell was in the first response

It wasn’t a fluke.

That extra depth held for the whole session, response after response.

By the time we wrapped, I called it out and asked Claude straight up whether the difference was real. (Yes, I asked AI to explain why the AI felt smarter. The irony isn’t lost on me.)

It’s the same model I run in the web app. Opus 4.8.

A better answer shouldn’t have been possible.

And yet.

Same brain, cleaner memory

The model’s weights are identical.

What changed was the quality of context it was reasoning from.

A long chat slowly fills with ambiguity, then gets squeezed into a shorter summary to save room.

Each step inherits a slightly blurrier picture than the last.

Kinda like a game of telephone. 😂

You’ve probably experienced this. You fix the same detail 3 times, and a few messages later, the chat serves it right back to you, wrong again. 🤦🏻‍♂️

This is called ‘decay.’

In Claude Code (and Cowork), the work lives in files, not in the conversation. At every step, Claude reads from clean, verified documents rather than a fuzzy recap.

A fresh, accurate base, every time.

The conversation was just the steering wheel. The files were the car.

Less decay, less error piling up, steadier output from one step to the next.

Smarter is a fair word for it, even though the brain never changed.

It isn’t just a developer tool

Claude itself has told me before that Claude Code is for developers.

Fair, as far as positioning goes. It’s intended for coding projects.

It connects to your computer’s terminal, and you have to be comfortable with Claude accessing and creating files. As a safe default for most people, calling it a dev tool makes sense.

It also undersells what the thing is.

Strip away the code, and what’s left is Claude, a computer, and a place to save things. That’s it.

Nearly all knowledge work runs the same loop.

Research something, think it through, produce the thing, save the thing.

You don’t need to be an engineer to want that. Sign me up.

So the tool that could help you most has been wearing a sign that says it’s not for you.

Well. It’s for you, too, my friend.

How to use the app you already open this way

Here’s the part you can use today, no terminal required.

It all comes down to one idea.

Make files the memory. Keep the chat disposable.

Start with the instructions Claude reads before it works.

In Code, that lives in a file called CLAUDE.md. Picture a recipe card Claude reads before it begins to cook, so it always knows who it is and how you like things done.

YES, CHEF!

You won’t create that file in the web app, though.

You recreate its job in your Project’s (custom) Instructions, the box you fill out once when you set up a Project.

Drop in Claude’s role, your standards, and the one rule that matters most. Tell Claude to read what’s already here (in your project files) first, and don’t rebuild what’s written down.

Once you’ve added Instructions, every new chat in that Project automatically reads them.

Next is the piece that actually stops the decay.

My STATE.md template. Start with Status and Next actions

In Code, it’s a file called STATE.md, a plain ‘where we left off’ note. What exists, what you decided, what’s still open, and what’s next.

In the web app, you make that same note as a regular document and keep it in your Project’s knowledge base, the shelf that every new chat reads before it answers.

That knowledge base is just the files you add to a Project. In every chat inside a Project, Claude can see them.

Creating the note takes one sentence.

At the end of a work session (that spans multiple sessions or days), ask Claude to write a short status summary of what you covered and what’s next, then save it in the Project.

Updating it asks a little more of you than Claude Code does.

Claude can read your current note and hand you a fresh version, but in the web app, it can’t overwrite the file for you. So you add the new one, then delete the old one immediately.

A couple of extra clicks, and the decay you’ve been fighting is gone.

That shelf (your Project files) is the habit to build.

Your status note is the most important part, but it isn’t the only part.

Anytime a chat produces a keeper, whether it’s a plan, a draft, or a decision, put it back on the shelf, too, so tomorrow’s chat builds on it instead of starting from scratch.

Then keep each chat to one task instead of one sprawling mega-thread (the mega-thread is exactly where decay creeps back in).

Connect your sources, such as Google Drive and Notion, so Claude can read your real documents from where they already live.

Remember, you can ask Claude to help you with any or all of this. Just ask it like you would a friend.

The worry that dissolved

Export all of your chats and data here

One thing still nagged me.

I love that every chat in my account is saved and searchable.

My Code and Cowork sessions on the desktop aren’t, so a new laptop could wipe them.

Then it clicked. Everything worth keeping from those sessions is already in the files.

And my files had been synced and backed up to Dropbox the whole time.

The account chats I didn’t want to lose have a safety net, too. From Settings, then Privacy, you can export all of them at once and keep your own copy.

The chats are the disposable layer.

Losing the chat stops mattering the moment the files hold everything worth keeping.

That first response wasn’t magic, and it wasn’t a better model.

It was a cleaner desk.

I’m setting up my Claude web app Projects this way now. Files first, chat second.

Top-grading everything I’ve been doing, and fixing what I’ve been doing wrong.

That way, whether I’m using the web app or the desktop app, the output quality will be the same, and the work keeps its edge from one session to the next.

This won’t make a one-off question better. The payoff shows up in real projects that span days, not in quick questions.

Before your next big project, take a few minutes to write down who Claude is for you and where things stand. Let the files carry the memory.

What’s the project you keep restarting from scratch every time you open a fresh chat?

Hit reply and share with me.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

🛒 AI Is Changing How Customers Choose Your Business
Customers are asking AI which business to pick. Here's how to make sure yours is in the answer.

🎬 Meta Brings Back Facebook Creator Studio
Facebook's creator hub is back as an AI sidekick that even drafts your replies. Useful, with one catch.

🆔 How to Reserve and Change Your WhatsApp Username
WhatsApp usernames are coming. Claim yours now so customers can reach you without your phone number.

My Favorite Things 💙
What I’m reading, watching, and enjoying

📜 The American Experiment

My wife and I watched this 5-part series last week, and it was fantastic.

It traces how the United States was founded and by whom, timed to the country's 250th anniversary, with Martin Sheen voicing George Washington and voices from across the political spectrum weighing in. Directed by Brian Knappenberger and executive produced by Tom Hanks.

What stuck with me is how thoughtfully it connects the founding to the challenges we face as a nation today. I can't stop thinking about the parallels and how much we think has changed, even though so much has been this way since the very beginning.

Good morning. Happy 4th. The big 250. 🇺🇸

Last week, a little meeting prep turned into something I didn't expect.

I needed to do some research to show up ready to explore an opportunity.

By the time I was done, I had a full business plan I never set out to write.

That accident is today's issue. I'll share what happened, then hand you the tool I built from it.

Today, let's talk about:

• How one research session turned into an entire business plan

• The simple 3-step move that gets far better output from any AI

• The free AI Co-Founder prompt you can run for your own ideas

Let’s get into it.

— Gordon

🧭 How I Accidentally Built a Business Plan

I finally tried running a team of AI agents.

I only meant to do some research.

They didn't stop there.

Here’s what happened. Last week, I sat down to prepare for a meeting.

Two friends built a product business in 2021, but never launched it. Life got busy, priorities shifted, and it sat.

Now they're dusting it off, and they've asked me to come in as a potential third co-founder.

I opened Claude to research a few things before we talked. The industry, competitors, and risks. I also wanted to review what they'd already built.

90 minutes later, I had an entire company's worth of research and strategy I never set out to build. (For a consultant who preaches scope, that's a little on the nose.)

How a research session turned into a business plan

Instead of dumping commands into a chat, I came from curiosity.

I asked Claude how it would approach the whole thing before it did anything.

The simple prompt that started it all 💁🏻‍♂️

It gave me 4 options and a recommendation.

I gave it a bit more context, and from there, Claude wrote 4 prompts, one for each research agent.

Most would run with that, but I was still curious and asked, "What other agents would be best for this? I don't know what I don't know. What should I be learning, researching, doing, considering, and all the things?"

Not a technical, Opus 4.8-optimized prompt, I know.

But it gave me prompts for 4 more agents, along with the most important thing: 12 questions to ask the other co-founders.

After a bit more back-and-forth, Claude started to cook. 🧑‍🍳

It drafted a thorough plan, tailored to this exact business.

Then it generated an instruction file (CLAUDE.md) for Claude Code to work from.

What happened next is the part I keep thinking about.

It didn't answer in one big block of text in the chat. It worked like a team.

A market researcher went first.

Then, a competitor analyst read the market research before it started.

Then, legal, operations, a financial model with live formulas, channel strategy, a brand audit, a growth plan, and a go-to-market strategy. Oh my! 🤯

Each one picked up where the last left off.

This compounded the work rather than repeating it.

The output was incredible, so I kept feeding it. Look at this, too. Now research that.

When the dust settled, there were 13 documents, and they read like one plan rather than 13 disconnected chats. Each doc even had data sources linked at the bottom.

Claude cooked 😮‍💨

Every doc had a TL;DR up top, so I read the whole stack in one sitting.

The most important thing I had Claude research and draft was a founder-structure doc. It maps out how to divide roles and ownership fairly.

Clarity protects a friendship, ambiguity destroys it.

None of it felt like using a chatbot anymore.

I sat down to prep for a meeting and stood up with a plan to launch a company.

It all came down to one move.

The move you should steal

I’ve said this before, and it’s worth repeating.

AI does the work. You do the thinking.

1. Tell the model what you're trying to achieve

2. Ask it for the best way to get there and which tools to use

3. Inject your own thinking to ensure it delivers exactly what you need

That works on almost any AI task, not just building a business.

The last step is where most people stop. They take the first answer as the final answer.

Staying in it is what caught the one thing Claude skipped.

The question AI skipped

It’s up to you to point out what AI missed 💅

For all that firepower, AI had one blind spot.

It assumed the idea was worth building and went straight to the how.

Validation was in there. Sort of.

A pre-order that tests whether people will pay.

An audience to build before spending a dollar on inventory.

But it never led with the real first question.

Should we even do this, and what's the cheapest way to find out?

Since I planned to templatize this prompt for other people, I wanted to make sure validation was included. It wasn't, at least, not as its own step.

So I made it step one in the version I'm handing you.

Validate before you plan. Cheapest tests first. Talk to real people.

See if anyone's already searching for this. Find the smallest signal before you pour hours into a roadmap.

A folder of confident documents can feel like a validated business. It isn't one.

Everything in there is a smart set of assumptions to test.

The idea becomes real the day a stranger pays for it.

I turned the whole process into a prompt you can run yourself.

It makes you test the idea cheaply before it builds you a plan.

Run it on the idea you've been sitting on.

Here’s the link: AI CO-FOUNDER PROMPT

Be safe this weekend. 🎆

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

⏳ When Developing an AI Strategy, Beware the Urgency Trap
Why most AI projects stall, and the shift in thinking that separates the ones that work.

🕶️ Meta Adds Custom Instagram Stories Features for Glasses
The camera on your face is turning into a content studio. Meta keeps proving it.

📉 Here's What Chief Economists See for a Bumpy Year Ahead
Nearly 9 in 10 economists see weaker growth ahead. Worth a peek as we enter the second half of the year.

Good morning.

Last weekend, I built my wife a Claude Skill that hands her back an hour every Sunday. It automates a chore she'd done manually for years.

Building it taught me more about creating AI Skills than any setup guide has.

Today, let's talk about:

• How to turn your workflows into a reusable system

• Three curated articles on where AI is headed

• A book that made this overthinker feel seen

Let’s get into it.

— Gordon

🔁 How to Build an AI Skill You Can Trust

On Sunday, I built my wife a Claude Skill that hands her back an hour every week.

It works for one reason, and it has nothing to do with how smart the AI is.

The common advice is to spin up a Skill for whatever task is bugging you.

So people build one from a rough idea, run it once, see output that looks acceptable, and ship it without ever testing it or defining what good looks like.

The Skill produces inconsistent output that the builder can't easily diagnose, so they end up guessing what the instructions should say instead of knowing.

A Skill is only as good as the process you feed it.

The leverage isn't the Skill feature. It's the documentation and automation of a process my wife had already run by hand for years, with the same cadence and format, yielding a known-good result.

All I did was write it down and teach Claude to run it her way.

That's the part the setup guides skip.

Everyone can create a Skill now.

Almost nobody tells you the prerequisite is a process you already trust.

HERE'S HOW:

• Pick a Process You Already Run by Hand

• Document Every Input and Step

• Define What "Good" Looks Like

• Test It Before You Schedule It

• Install It and Walk Away

Here's what each step looked like in the build and how to run it on your own process.

Pick a Process You Already Run by Hand

The best first Skill automates a ritual you already run reliably.

The same workflow, the same result, every time.

It's tempting to build a Skill for the messiest, most annoying task on your plate.

Unfortunately, that's the one least ready to automate, because you've never pinned down a repeatable way to do it.

The meal plan qualified because it had been a Sunday ritual for years, with an output we already knew was right. There was nothing to invent. The process existed, it just lived in someone's head and an Apple Note.

The test is simple.

If you can't do it the same way twice by hand, it's not ready to automate.

If you can, you've got your candidate.

Document Every Input and Step

Write down exactly what the process reads and what it does, in order.

For example, the meal planner takes 3 inputs.

Our family Google Calendar (so it knows which nights are busy), the week's weather for our ZIP, and a Google Doc of our go-to recipes.

We end up with a finished plan, a meal for each day, and grocery lists split by store.

List the inputs first, then the steps that touch them. If you can't define an input, Claude can't execute on it.

Vague notes in, inconsistent Skill out.

People rush this step. It's the one that decides whether the Skill works at all.

Define What "Good" Looks Like

Claude matches the target you give it. Make that target specific.

The plan needed the grocery list broken out by where we shop.

It needed a weather emoji on each day, the exact format that used to live in Apple Notes. That detail mattered more than I expected. It's what keeps the plan feeling handmade instead of generic.

The fastest way to define good is to hand Claude one finished example you'd be glad to receive, then tell it to match the shape. I simply gave Claude a screenshot of our current Apple Note.

Show Claude a model of done, and it stops guessing.

Skip this, and you get something that's technically correct, yet just not the same.

Test It Before You Schedule It

Do it once, and inspect what you expect, before you trust it to run on its own.

We did, and the dry run caught what the plan on paper missed.

The recipes were buried in the same doc as the instructions, so updating the menu later could break the workflow. We split the recipes into their own Google Doc, and now the list can grow for years without anyone touching the Skill.

Test it on a normal week and a strange one.

The normal week proves it works. The odd week, a holiday, or an empty calendar, shows you where it breaks.

Always be testing, my friends.

Install It and Walk Away

Now you turn it into a Skill and schedule it.

Our Weekly Meal Plan Skill now runs every Sunday at 7 AM. It reads the calendar, checks the weather, pulls the recipes, writes the plan into a fresh Google Doc, and drops the link in Slack when it's done.

You can walk away only after you confirm it works. That's what the test bought you. Automating an unproven process just schedules the mistakes.

Same inputs, same quality, no hand-holding.

That's the leverage, and it works because the 4 steps before it were proven.

This coming Sunday at 7 AM, our meal plan will be delivered as we're waking up.

All we'll have to do is confirm no changes are needed and check the fridge and pantry to finalize our grocery list.

An hour a week is about 50 hours a year, and it came from getting one process right, then documenting it so that AI can repeat it.

Your best first automation works the same way. It's something you're already good at, handed off.

If you can't do it the same way twice by hand, it's not ready to automate. If you can, you've found your first Skill.

Hit reply and tell me the one task you'd hand off first at work or at home.

See you next week.

ONWARD TOGETHER.

Next Steps 👇

PS — Whenever you're ready, here's how I can help

A free 30-minute discovery call. We'll pinpoint the 1-2 places in your unique business where AI would save you the most time, so you leave with a clear starting point. If you want help building it out, I'll show you what that looks like.

Everyday Insights 👀
A few things worth reading this week

📱 Facebook rolls out an AI companion app for creators
Meta is building creators their own AI sidekick. Worth watching where this goes before everyone has one.

🎨 LinkedIn lets users set core brand rules
LinkedIn's letting you hand its AI your brand rules. Quietly a big deal if you post there.

📊 The 5 Types of AI Investment and How to Capture Their Value
A framework for AI spending that ditches the usual ROI math. Read it before your next tool decision.

My Favorite Things 💙
What I’m reading, watching, and enjoying

🎧 Soundtracks: The Surprising Solution to Overthinking by Jon Acuff

I flew through this one on Libby last week (free library audiobooks still feel like a cheat code).

As a lifelong overthinker with ADHD tendencies, I felt seen immediately. The reframe at the center of the book is that overthinking is fear in disguise, the gap between what you think and what you actually want.

His fix is refreshingly simple. The thoughts you replay are "soundtracks," and you change them in three moves. Retire the broken ones, replace them with better ones, then repeat the process with the new ones until they run on autopilot like the old ones did.

If your brain is like mine and talks you out of things before you start, give it a listen.

Good morning.

Every few weeks, a new term in the AI world pops up that sounds like it requires an engineering degree. Plenty of them are noise.

This week, 2 of the tools I use every day added one that's worth your attention.

So I put it to work and wrote up what it means for a business like ours. The name sounds technical. What it does is simple, and it can hand you back hours.

Today, let's talk about:

• The 3 letters showing up everywhere

• Why every app you use is racing to add it

• How to put it to work today, with prompts to steal

Let’s get into it.

— Gordon

🔌 How 3 Letters Can Take Work Off Your Plate

I first learned about MCPs a year ago on Greg Isenberg's "The Startup Ideas Podcast."

Since mid-April, I've gotten at least one email a day with creators talking about MCPs.

Then on Monday, the Eden app I use launched its MCP, and on Tuesday, beehiiv released a major update to its own, adding write capabilities.

I know I'm speaking gibberish here, but things are becoming very interesting in how you can use your AI of choice with your favorite tools.

MCP is the connector that quietly lets AI take real work off your plate.

Plain English, please

MCP stands for Model Context Protocol. In plain terms, it's a shared standard that lets your AI plug into an outside tool and get things done inside it.

Think USB-C for AI.

Before this, every connection between an AI and an app was a custom, one-off build, or adding a Zapier or Make subscription to your P&L.

Now there's one universal port.

Any AI that speaks MCP can connect to any tool that speaks it.

Why it's suddenly everywhere

One shift did it. AI stopped just answering questions and started doing the work, and an assistant that can't reach your tools is only a smarter chat box.

So every platform is racing to ship an MCP and stay useful inside the AI you open every day.

And it happened fast.

Anthropic introduced MCP at the end of 2024. OpenAI added it in early 2025, Google and Microsoft followed by the summer, and by December, the rivals who compete on everything else agreed to govern it together.

When competitors standardize on the same thing, you know it's here to stay.

The gap is about to crack wide open. While you're stuck, drowning in admin work, your competitors with MCPs connected can focus on generating more business.

How to actually get something out of it

The payoff shows up when you connect the 2 or 3 tools you live in and stop copy-pasting between them.

Tuesday afternoon, I connected Eden, my content research tool, and asked Claude to dig up the best-performing short-form videos in my niche and draft a script off the winning structure.

I went from "let's test this sucker" to a finished script in 20 minutes.

No browser tabs galore, no copy-paste, no hours spent deep in the rabbit hole.

Yesterday, I asked Claude to pull my latest newsletter numbers out of beehiiv and log them in Notion.

It read straight from beehiiv, built a clean Notion database, named the columns, and entered the open and click rates for each issue.

One ask, done.

It was done so fast that I had Claude backfill the database with all of my newsletters. I went for my daily walk, and it was finished when I got back. 🤌

I wouldn't have known how to complete that task, but Claude accessed the knowledge bases for beehiiv and Notion, determined the best workflow based on the tools available using each app’s MCPs, and executed it.

That's 2 tools handing off within a single request, no export, no copy-paste.

Add a third, and the handoffs keep going, all from a single message, while you go do something else.

Here's the part most people skip.

Bolt AI onto a workflow that's already broken, and all that does is speed up the mess. Fine-tune or fix your process first, then connect the tools that serve it.

The temptation is to connect everything, because you can. Don't.

Connect what you'll actually use. The sidebar isn't a trophy case.

Again, connect what matches a real workflow and ignore the rest.

It's the safer play, too. Every tool you wire in is one more door into your stuff.

beehiiv rolled out its MCP in phases on purpose, read-only first, write second, so you stay in control of what goes out.

The Daily Driver Rule

A tool earns its spot the day you've got a job for it.

Before you connect anything, ask 2 questions.

1. Is this a daily driver, and something I actually live in?

2. Is there a job I'd hand it today that I dread doing?

2 yeses, wire it up. Otherwise, leave it.

Start with one, then steal these

Connecting a tool takes a minute. Like, one.

In your AI of choice, open the connectors settings, switch on the one you'd be lost without, and authenticate. In Claude, it's in Customize > Connectors. 🧰

Begin with one tool, then see what happens when 2 start talking.

Here’s a single-tool win for your Sunday planning. Connect Claude or ChatGPT to Google Calendar or Microsoft 365 on read-only, copy and paste this prompt into a new chat, and let ‘er rip.

I ran that prompt in Claude, and it COOKED. 😮‍💨

Here’s a cross-tool flow for one ask across your inbox and a spreadsheet. You’ll need to connect Gmail and Google Drive on read-only for this one.

Start with the tedious job you repeat every week.

Prove it saves you time before you connect another app.

A year ago, I first heard the term MCP on a podcast.

This week, 2 of my own tools shipped one.

By the end of this summer, we'll be cooking with gas. 📈

Connect an app this weekend and hand AI a job you'd rather not do. Every hour it takes off your plate is an hour back for the work only you can do.

See you next week.

ONWARD TOGETHER.

Next Steps 👇

PS — Whenever you're ready, here's how I can help

A free 30-minute discovery call. We'll pinpoint the 1-2 places in your unique business where AI would save you the most time, so you leave with a clear starting point. If you want help building it out, I'll show you what that looks like.

Everyday Insights 👀
A few things worth reading this week

🧠 To Thrive Alongside AI, Focus on Mindset—Not Skillset
Everyone's asking which 10% of their job AI can't touch. A Goldman Sachs CIO argues that's the wrong question, and the better one changes how you think about your edge.

🎬 Edits is getting a desktop version and an AI production assistant
Meta's Edits, the app many creators use to cut Reels, is adding a desktop version and an AI assistant that does more than just trim clips. If short-form video is on your list, take a look.

🎛️ Social media's next evolution: User-controlled algorithms
The major platforms are starting to let you train your own feed, telling the algorithm what you want more and less of. Worth understanding from both sides, as someone who scrolls and as someone whose content needs to get found.

My Favorite Things 💙
What I’m reading, watching, and enjoying

🧀 Who Moved My Cheese? by Spencer Johnson

I read this one in an hour this week, and my first thought was that I wish I'd picked it up when it came out almost 30 years ago.

It's a short parable set in a maze, where 2 mice and 2 little people live off a big pile of cheese. One day, the cheese is gone. The whole story is about how each of them reacts. Some go looking for new cheese right away, while others freeze and wait for the old pile to come back, complaining about how inconvenienced they are instead of jumping into action to find a solution.

The lesson is simple. The cheese (i.e., your ambitions and goals) always moves, and the people who watch for change and adapt early spend a lot less time stuck than the ones who insist things should stay the way they were.

Feels especially true right now, with how fast so much is changing, all at once.

Good morning.

For years, "just ship it" was the best advice in content. I think it just expired.

AI is the reason, and not in the way you'd expect.

Today, let's talk about:

• What happened to the script I told you was the one

• The script that read great and sounded wrong

• The one test that catches "fine"

Let’s get into it.

— Gordon

🪞 How To Use AI Without Becoming Slop

A few weeks ago, I showed you the script for my first Reel.

I was ready to film it.

Then I read it through one more time and quietly cut it.

It wasn't bad. It just wasn't great, and it wasn't me.

This was the first thing anyone would ever see of me on camera. It had to be good, and it had to sound like me.

The script I'd already committed to

The “102 lead magnets” script went out in a newsletter. I told you it was the one.

Then I realized it wasn’t written for how I wanted to make content.

I wanted the first Reel to feel like we'd grabbed a coffee together to chat.

Calm, conversational, and using my inside voice.

Not those Reels with lightning-fast pacing, high-energy yelling, and constant cuts that make your brain short out. 😵‍💫

I also wanted it to be closely tied to the work I do and the people I serve.

The script was honest and competent, but it wasn't right.

When my ears caught what my eyes missed

On the couch. Just me and my iPhone.

I wrote a new script about how AI can now take half of the 170 tasks off a real estate agent's plate.

On the page, it was good. Better than the first, for sure.

I loaded it into the teleprompter in the Edits app and recorded a rough take on my couch to hear it out loud.

No light, no mic, no setup. Just me and my iPhone.

It didn't sound like me. It came out stiff and performed.

Nothing like the casual conversation I was going for.

So I cut that one too.

The taste test

Hold up, wait a minute, something ain't right

After the second cut, I realized there was a pattern.

I call it “The Bookend Method.”

I go first, AI builds with me in the middle, and then I make one last pass.

Both scripts came from real things I'd done, so the raw material was already mine.

The last pass is the Taste Test. A quality and authenticity check.

Catching AI when it fabricates is easy. Everyone does that.

The Taste Test isn't rejecting what's false. It's rejecting what's fine.

They were both accurate. Both were usable. I killed them because they were the competent, but generic, version of me, and I wanted the real me.

The fastest way I've found to catch that, for anything you're going to say out loud, is to actually say it out loud and listen for the moment you stop sounding like yourself.

I've always done that with my newsletters, and I’ve learned it's even more important with video.

Why "just ship it" stopped being enough

Fine? You know what "fine" stands for, don't you?

“Progress beats perfection” was great advice for a world where making things was hard. AI removed the hard part.

Anyone can generate “fine” in 10 seconds now, which is exactly why “fine” has turned into a liability. People can feel the machine, and they're voting with their attention.

More than half say they pull back the moment they suspect something was AI-made, and 2 in 3 now say they're choosier about what they engage with than they were a year ago.

Instagram's Adam Mosseri has gone so far as to call authenticity a scarce resource, where the real bar is whether you can make something only you could.

When competence is free, and trust is scarce, what sets you apart is the nerve to turn the polished thing down and hold out for the one that's yours.

That's taste, and it's the one part of the work nobody can generate for you.

So here's where I am today.

I've cut 2 scripts, I haven't written the next one, and I still haven't posted a single Reel. 🫠

The uncomfortable truth is that taste and avoidance look the same from the outside, and from in here, I can't always tell whether I'm protecting a standard or hiding behind one.

What I do know is that when AI makes “fine” easy, your taste becomes the edge, and the real work is learning when taste turns into hiding.

What's the last thing you shipped that was just fine, and what are you holding back that's actually you?

Hit reply and share with me. I'm still working out the line myself.

See you next week.

ONWARD TOGETHER.

Next Steps 👇

PS — Whenever you're ready, here's how I can help

A free 30-minute discovery call. We'll pinpoint the 1-2 places in your unique business where AI would save you the most time, so you leave with a clear starting point. If you want help building it out, I'll show you what that looks like.

Everyday Insights 👀
A few things worth reading this week

🏷️ Why Effective Leaders Get Branded as Problems
When a leader creates friction, everyone assumes the leader is the problem. This makes the case that it usually traces to one of 4 hidden sources, and misreading which one is how companies lose their best people.

💸 Instagram Outlines Add-On Subscription Offerings
Instagram just launched its first paid tier, Instagram Plus, at $3.99 a month. Here's everything you get, plus the case for why most of it isn't worth paying for yet.

🦋 Claude Fable 5 and Claude Mythos 5
Anthropic just shipped Fable 5, its most capable public model yet, and its lead grows the longer and harder the task gets. It's free on the paid plans through June 22, so it's a good week to throw your toughest work at it.

My Favorite Things 💙
What I’m reading, watching, and enjoying

📚 The Millionaire Next Door by Thomas J. Stanley and William D. Danko

I've been listening to this book and will finish it this week. It's dated, and the first half is a slog of survey data that bored me to tears. The second half is where it pays off, finally, clarifying how consumer spending and real wealth actually work.

The big takeaway is that wealth comes from what you keep and how consistently you invest it, so the people quietly building it tend to live below their means at any income.

Good morning.

A lot of what I share here, I'm working out in real time.

This issue is a good example.

Yesterday, I rebuilt something I should have fixed ages ago, and I documented it as it happened. It's about idea capture, why systems quietly die, and the one change that finally made mine stick.

If you've ever set up a system and then quietly stopped using it, this one's for you.

Today, let's talk about:

• The idea list I never opened

• The rebuild I didn't see coming

• The one reason this system works

Let’s get into it.

— Gordon

2026 State of AEO Report

A year ago, most marketers weren't thinking about AI search. Now it's one of the fastest moving channels in the industry and nobody has a playbook yet.

So we built one. We surveyed hundreds of marketers to find out how they're approaching answer engine optimization, where they're investing, what's actually working, and what isn't.

The result is the 2026 State of AEO Report. Real data. Real strategies. A clear picture of where AI search is headed and how to get ahead of it.

Download the free report

🪴 How I Finally Built A Capture System

I've told you before to build a capture system.

A home for your ideas so you never sit down to a blank page.

I had one. I just wasn't using it.

That's the uncomfortable part of teaching what you’re learning.

You can know the right move and still not make it.

Back when I started this newsletter, I set up an idea list from a template in Justin Welsh's old ContentOS course.

A clean Notion checklist with one job.

Capture newsletter ideas to get the ball rolling.

A simple system for a more civilized age

It was a good system, parked in a corner of my workspace, which I rarely opened.

So I rarely fed it, and I almost never checked it before writing.

Every so often, I'd go back to it, hunting for something to write. The list was stale.

Worse, the ideas were too thin to use. One just said "Legendary Lead Gen."

I have no idea what it means or what I was going to write.

I’m sure it would have been legendary, though. 😂

The spark was real once. A few words on a list I never revisited couldn't keep it alive.

A capture system I couldn't build from wasn't helpful. It defeated the point.

The fix showed up by accident

Yesterday, I wasn't planning to solve this.

It wasn't on any to-do list. I was deep in other work when it clicked that my capture system could be implemented differently, and I decided to rebuild it right then.

I knew I wanted a real database this time, shaped around the way I actually work.

What I didn't know was which columns it should have. I’m not a spreadsheet guy.

So I asked Claude to recommend some.

It gave me 7 core options and 5 optional adds.

When you don’t know what you don’t know, just ask Claude

It suggested skipping a couple to keep things lean. Yolo, I added them all.

I know how I work, and I'd rather grow into the columns (or remove them) than wish down the road that I'd built them in the first place.

Not usually the best approach, but that’s how I work.

Claude built the database in Notion, right where the rest of my newsletter systems live, and gave me the direct link to review its work.

Why this system will thrive

Nice and tidy, just the way I like it

The old list died because it sat off to the side, outside the flow of how I write.

This one lives inside that flow. I can drop an idea as soon as it shows up. Claude can also add ideas while we work together.

When it's time to choose the next piece, Claude reads the database as part of its research to find current, validated ideas to inform my decision.

A system only works if it lives where the work already happens.

Same kind of list, a completely different result, because this one is wired into my process instead of parked elsewhere.

Keep your idea list inside your daily workflow. Anywhere else, it dies.

That ContentOS template is still in my workspace. I'm not deleting it.

It taught me a lesson the slow way.

A system you build and then abandon isn't a system.

It's an expensive souvenir.

See you next week.

ONWARD TOGETHER.

Next Steps 👇

PS — Whenever you're ready, here's how I can help

A complimentary, 30-minute discovery call. We'll pinpoint the 1-2 places in your unique business where AI would save you the most time, so you leave with a clear starting point. If you want help building it out, I'll show you what that looks like.

Everyday Insights 👀
A few things worth reading this week

🧠 Big Tech's Looming Capability Crisis
AI made output cheap. Here's the case for why the human part just got more valuable.

🎬 Instagram Introduces Teleprompter Tool
Instagram moved its teleprompter into the main camera. After this week, I've got some thoughts.

🔓 Instagram Is Alerting Users Targeted By AI Chatbot Hackers
Hackers took over accounts by politely asking Meta's AI. It actually worked. 😅

Good morning.

Anthropic (Claude.ai) shipped something two weeks ago that you may have missed. Today, we’re walking through the orientation and the on-ramp.

We’ll talk about:

• What just shipped

• Why it matters for you

• Crawl, walk, run: how to start

Let's get into it.

— Gordon

Talk to your AI tools the way you'd talk to a colleague.

You don't send a colleague a three-word brief. You explain the context, the constraints, what you've already tried. But typing all that into ChatGPT takes forever — so you don't.

Wispr Flow lets you speak your prompts instead. Talk through your thinking naturally and get clean, paste-ready text. No filler words. No cleanup. Just detailed prompts that actually get you useful answers on the first try.

Millions of users worldwide. Works system-wide on Mac, Windows, and iPhone.

Try Wispr Flow free

🧰 How To Use Claude For Small Business

Anthropic just launched Claude for Small Business.

If you pay for Claude Pro, Max, or Team, you're already sitting on 31 prebuilt skills ready to cook.

Toggle one switch, and you’re Tony Stark.

Tony Stark aka Iron Man

The launch was two weeks ago and didn't get much coverage outside AI Twitter and Reddit. Here's what it is, why it matters, and how to start using it this week.

What's included

The Skills to Pay the Bills

Claude for Small Business is a plugin you toggle on inside Claude Cowork.

It ships with 31 prebuilt skills covering finance, operations, sales, marketing, HR, and customer service.

It connects to 12 tools you probably already pay for: QuickBooks, PayPal, Stripe, Square, HubSpot, Canva, DocuSign, Slack, Microsoft 365, Gmail, Google Calendar, and Google Drive.

You pick which to connect (one at a time, via the standard OAuth flow), so you're not handing over your whole stack on day one.

The named target verticals at launch were a 15-person HVAC company, a 30-person landscaper, and a 50-person real estate brokerage.

So if you're sitting there wondering whether this is for you, the answer is yes (especially if your business looks anything like one of those three).

Every skill waits for your approval before anything is sent, posted, or paid.

Anthropic also launched a free AI Fluency for Small Business course alongside the plugin, taught by actual small business owners.

Why this matters

The hard part used to be the starting question.

Where do I even use AI in my business?

You'd read 12 articles, bookmark 4 newsletters (obviously this one 😉), watch 3 YouTube videos, and still have no idea what to do.

Claude for Small Business solves that.

You don't need to figure out where AI fits in your business. Anthropic just gave you the whole toolbox.

Think of it as a starter pack of the 31 skills most small businesses need, built to plug into the tools you already pay for.

Even the starter pack is shapeable.

Click "Customize," and Claude tailors the skill set to your specific business.

The real question now is how deep to go and how fast.

Crawl: Run one skill once

Claude desktop app

Toggle on Claude for Small Business in Cowork.

Go to Customize, Browse plugins, then scroll to find Small Business by Anthropic. Either click the + or click the card to open it, and then click Install.

Connect QuickBooks and HubSpot first.

Run the business pulse skill. It's a one-page brief.

Cash position. Sales trend. The deals most likely to close this week. The commitments that are waiting for you.

One ask, one output, immediate proof that this is real. The business pulse is the skill I'm most curious to test first.

Of course, this assumes you’re using those apps. If you’re not, connect one or two that you are using and run a related skill.

Walk: Plug the biggest leak

Crawl was about seeing the thing work.

Walk is about pointing Claude at the leak in your business.

For most service businesses, that leak is speed-to-lead.

You pay for the lead, then take 4 hours to respond, then wonder why your close rate is zero. “The leads are weak,” you say.

The lead triage skill hooks into HubSpot (the only CRM connector at launch).

It sorts incoming leads, scores them based on fit, and drafts the initial reply.

You still approve before anything sends. But the response goes out almost immediately, not two days later.

Speed-to-lead research is crystal clear.

Leads contacted within 5 minutes are 21x more likely to qualify than leads contacted at 30 minutes. With first-reply drafts ready in minutes, you actually hit that window.

If marketing is your bigger leak, swap in the campaign skill instead.

It finds slow stretches in revenue, analyzes what worked in HubSpot, drafts a strategy, and generates Canva assets.

Run: Make it the operating layer

/monday-brief Skill

Run is where Claude for Small Business stops being a tool you visit and starts running underneath the business.

Lead triage handles first-touch. The follow-up sequence runs from there.

The business pulse keeps you informed.

The campaign skill feeds new top-of-funnel.

You stop being the bottleneck.

You stop doing the after-hours catch-up work.

Your week is starting to look more like a business owner's.

Two weeks ago, this didn't exist. Today, there's a path in at any speed.

Don't overthink this. Just open Cowork, flip the toggle, run the pulse.

See you next week.

ONWARD TOGETHER.

Next Steps 👇

PS — Whenever you're ready, here's how I can help

A free 30-minute discovery call. We'll pinpoint the 1-2 places in your unique business where AI would save you the most time, so you leave with a clear starting point. If you want help building it out, I'll show you what that looks like.

Everyday Insights 👀
A few things worth reading this week

🧪 How To Move From AI Experimentation To AI Transformation
Adoption is easy. Real transformation is rare. What makes the difference?

📸 Instagram Engagement Rates Provide Insight Into Reach
The metric driving Instagram reach, straight from the platform's head.

💬 Meta Launches New Group-Focused Forum App
Meta launched a Reddit clone for Facebook Groups. The why is more interesting than the what.

My Favorite Things 💙
What I’m reading, watching, and enjoying

📗 Profit First by Mike Michalowicz

Mike flips the accounting formula.

The standard one is Sales - Expenses = Profit, which leaves profit as whatever's left over (often nothing). Profit First runs Sales - Profit = Expenses.

You take profit first, then live on what's left.

The system runs on multiple bank accounts. Income, Profit, Owner's Pay, Tax, Operating Expenses.

When revenue hits, you allocate percentages to each. The constraint forces discipline, because you can only spend what's in the Operating Expenses account.

It's basically Parkinson's Law for money. Expenses expand to fill the money available, so you shrink the available money and find ways to make it work.

I'm planning to adopt it this year while I've got a relatively clean slate. The system seems built for solopreneurs and small business owners running lean.

Mike narrates the audiobook himself, and his subtle humor keeps a dry topic moving. Mercury's breakdown is a solid primer if you want the short version.

Good morning.

Last Thursday, I shared my Reels workflow. By Tuesday, I'd bypassed it. The recovery turned into a framework.

Today, let's talk about:

• The shortcut I caught in real time

• Script #1: 102 lead magnets. Zero results

• When to start with a story vs. finding an outlier

Let’s get into it.

— Gordon

🤝 Where Story And Format Meet

Tuesday afternoon, I landed on a script for my first Reel.

Then I started thinking about the editing.

The thought of doing this multiple times a week, every week, was super overwhelming.

I opened Claude, looking for a shortcut. I asked for three Yap scripts based on what I'd done this week.

Talking-head, casual, low-edit. Anything to make the editing feel manageable.

The trick is, I'd just published a newsletter last week about the method I believe in for researching and validating Reels.

Less than 24 hours later, I'd already bypassed the workflow I told you to use.

The shortcut created temporary relief

The Yap format requires minimal editing.

It also happens to be the most popular format in 2026.

Less editing meant less overwhelm. Just focus on the message.

The trade made sense. In the moment.

I'd been imagining the editing for a "normal" Reel. An Adobe Premiere Pro timeline that looked like a blockbuster movie.

Cuts galore, B-Roll, sound design, and the time suck of doing that 3-5 times a week.

Kind of like this old timeline of mine. A cool, 30-hour edit. 😅

Yapping felt like a reasonable cheat code.

Just talk to the camera, cut the dead air, and post. Boom. Done.

Three scripts in, I had something I could film tomorrow. It felt productive.

I closed the laptop and convinced myself this was fine.

The next morning, I woke up feeling like something was off.

The scripts were fine. The route I took to get them was the problem. I'd asked Claude for a shortcut around the finding proven outliers step from last week's newsletter.

I'd written about the method. The first chance to use it, I'd skipped it.

What I tried yesterday morning

I picked Script #1 of the three and tried to run it through Sandcastles.ai* (my content research tool) for a fresh rewrite.

Kind of like getting a second opinion. 😂

My overthinking was to triangulate three drafts (Claude's draft, my version, and Sandcastles' attempt) and synthesize the final.

Sandcastles didn't rewrite. It generated something new from scratch, ignoring most of what I'd written, and fabricating its own story. None of which was true.

Ugh.

Sandcastles is built for ideation. Asking it to retrofit was my mistake.

That's when I realized I was overcomplicating the process.

Two clear workflow options

I was overcomplicating things just to shore up my confidence.

There are two clear workflows here, and I was reaching for a tool built for the wrong one.

1. Ideation (no story yet): Sandcastles, start to finish. Find relevant outliers in your niche. Extract the structure. Research to support the idea. Write the script.

2. Story in hand: Match your lived story to a proven post format (Yap, clone, POV, green screen, multitasking, etc.). Flesh out the script with Claude.

Quick check: if you can say the lesson in one sentence, you have a story. If you can't, start with outliers.

This week, I should have just used Workflow 2. I had the story already. The 102 lead magnets analysis, the forgotten flagship course, and the lesson I wanted Reel viewers to walk away with.

Trying to validate it through Sandcastles was reaching for a tool built for Workflow 1.

The script I'm rolling with

Here's where I landed with my first Reels script.

It’s not a true Yap script, but it’s authentic and to the point.

Script #1: 102 lead magnets. Zero results.

That's Workflow 2 when I don't get in its way.

My story. A format that's proven. Every line earns its place.

The format provides the scaffolding. Your life fills it.

The real method is simple. Match the tool to the moment you're in.

If you have a story, the format provides the scaffolding to land it.

If you don't have a story, outliers give you a proven idea to write one.

Yesterday, I was overwhelmed by the complexity of the editing and asked Claude for a shortcut.

Today, Script #1 is locked and based on what worked this week.

The story is still mine. The format is proven.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

🏀 How To Build A Superteam That Keeps Getting Better
What changed when the NBA's worst team started getting dramatically better.

✂️ Meta Reveals Latest Round Of Edits Improvements
If you're editing Reels in Edits, these new updates matter.

🔍 Social Media Drives Consumer Discovery, Per Meta
Where consumers actually search now that Google searches dropped 20%.

Good morning.

Spent the week prepping for my first 10 Reels and learned more than I expected.

Today, we got:

• What 85 outliers taught me about my first 10 Reels

• AI for SMBs, leadership power skills, and LinkedIn for consultants

• The calmest AI walkthrough I've watched in months

Let’s get into it.

— Gordon

📊 What 85 Outliers Taught Me About My First Reels

I sat down this week to decide what to shoot for my first 10 Reels.

After avoiding video content for over 6 years, sitting down to pick what to shoot should have been easy.

Hundreds of analyzed videos in my niche sat in front of me, the highest-performing content in the space.

Not one of them felt like me.

Idea generation is overwhelming work. It’s super easy to overthink this part.

And oftentimes, the good ideas are already in your own experience, but they don't come to mind when you need them.

Top 100+ outliers by the top creators in my niche

There's a reason none of those outliers felt like me. It has to do with what "cringe" actually means for me. When creators talk about cringe, they mean tolerating posting bad content before you get good.

My concern about cringe is different. It's about authenticity.

I know my first few dozen videos will suck and feel cringe, but copying what works for everyone else and not finding what’s me is what would really feel cringe.

So the filter I built this week had two parts. Does this work, and is this me?

Letting the data guide me

The first part of that filter needed data.

Monday, I started running outliers through Claude using Sandcastles*, the analysis tool I'm using to ensure my efforts aren't wasted on shit ideas.

After analyzing a dozen videos, I thought I'd spotted a pattern.

And by “I spotted,” I mean Claude spotted. 🤓

Question and list hooks appeared to be underused goldmines in my niche.

I didn’t stop there.

On Tuesday afternoon, I traded the spreadsheet for Instagram. Pure content consumption. After scrolling for an hour (or two), I could feel which formats had a texture that matched my voice.

When I came back, I expanded the analysis by giving Claude another, larger batch of outliers. By the time the sample grew past 80, a different pattern emerged.

I’ll see your lists and questions and raise you a secret reveal 💁🏻‍♂️

The winning formats were heavily used and had strong medians, with 'secret_reveal_breakdown' leading the pack.

The discipline was to keep working and let the data correct what I thought I knew.

Stopping after the first batch would have cost me 12 hours of shooting the wrong videos. Ah. No, thanks.

The third filter

The data answered the first part of the filter. I found what ideas are proven.

Then I sat with the 10 picks Claude produced.

Something was off.

The hooks read like demonstrations.

"I built a Claude agent that scans local market data," and "I broke up with ChatGPT and moved my entire real estate practice to Claude." Except I hadn't done either.

I hadn’t been clear enough about what I wanted Claude to do. I didn't tell Claude to ask me first how I'd authentically build on those 10 outliers. So Claude jumped ahead and created hooks from the data.

This is why you can’t be lazy with AI. It’s happy to fabricate shit to appease you, but that wastes your time.

Quick recap. The dual filter from Monday had two parts:

1. Does this work?

2. Is this me?

The work on Tuesday and Wednesday uncovered a third.

Have I actually done this?

Let’s call it the Authenticity Filter.

It sounds obvious, and this is a great example of where you need to use your thinking, judgment, and taste to push back on AI to ensure it gets it right.

Plus, I could go do those things and then create content around them, but that’s extra work I don’t have time for, and there’s plenty of experience I can already draw on. I needed to be more specific in my instructions and tell Claude when to bring me back into the process.

The work before the camera determines whether the camera matters.

Research was just the first step of the work.

Now, I have to go back and choose an angle, a contrarian take, and write the hook for each of those 10 ideas. Hooks based on real experience.

Then, I’ll write the scripts.

The camera finally turns on after all of that.

For any project that ends in a camera, a launch, or a publish, the leverage is upstream.

If you want to grow your business or build an audience through content, the best way to protect your time investment is to create new versions of proven posts in your niche.

Build on what's proven in your niche by remixing the core components.

Coming up with one idea and running with it won’t yield the results you’re looking for. You need 100 to find the best 20 to prioritize the top 5-10.

The first Reel posts at the end of the month.

My process will begin to click and become faster with repetition.

Where in your creative work have you been jumping to execution before doing the prep?

Hit reply and tell me what your version of "the work before the camera" looks like.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

🧠 Why Leaders Need "Power Skills"
A quieter take on why trust at work keeps eroding. The skill angle landed.

🏪 Anthropic courts a new kind of customer: small business owners
A bundle of Claude features for small business owners just shipped. I'm watching this one.

💵 LinkedIn will enable consultants to book business direct from their profile
LinkedIn Premium just made it easier to turn your network into paid sessions. Big shift.

My Favorite Things 💙
What I’m reading, watching, and enjoying

🎙️ Ex-Amazon AI Leader: In 1 Year, the Gap Between AI Users and Everyone Else Will Be Irreversible by Silicon Valley Girl

Watched this on Tuesday. Don't let the title spook you.

The conversation itself is the calmest, most practical AI walkthrough I've heard in a while. Allie K. Miller talks through how she runs her stack, builds a skill live on camera, and names the 3 context docs every solo operator should start with, all without the usual scare tactics.

Good morning.

Spent the week clearing out clutter. Organizing my desk, closing a gazillion browser tabs, and doing some long-overdue cable management. The kind of reset that should've happened a while ago.

Spring cleaning applies to both physical and digital spaces. 🌱

Today, let’s talk about:

• A confession I've been carrying

• What $12,000 of dormant gear cost me

• A book I highly recommend you read ASAP

Let’s get into it.

— Gordon

😔 Why I've Been Avoiding My Studio For 6 Years

I have a confession.

I've been putting off something I've genuinely wanted to do for 6 years.

It doesn't make any sense.

Why put off something I'm that excited about?

Well, what started as a simple obsession turned into something I overcomplicated.

How it started

In 2015, I stumbled on Gary Vaynerchuk's #AskGaryVee Show on YouTube. I was immediately hooked on the laid-back, conversational format.

Shortly after, I discovered my favorite YouTuber, Peter McKinnon. I loved his vibe, the look and feel of his videos, and I became obsessed with the idea of creating videos around my interests in a casual, authentic format like Gary's, but with the look of Peter's vids.

Fast forward to 2018: I stepped into a Director of Technology role I'd helped create at my real estate brokerage.

I started filming content on my iPhone, and after about a year, I bought my first camera (a Sony A7iii). I learned to edit in Adobe Premiere Pro, and as I went deeper into the craft, I fell down the rabbit hole of production gear.

Wanting better audio and a tighter shot turned into this 😅

Going into 2020, I had built a full content production studio in my office and was ready to go all-in on creating content for me and the agents and leaders in my brokerage.

Before I could get started, I was hired for a Regional Tech Trainer role to support 5,000 agents and 100 brokerage leaders across 6 states. Then, the pandemic hit.

Since then, my roughly $12-15K worth of gear and the skills I had built have been dormant, collecting dust. 😩

When I launched this newsletter in May of 2023, the goal was to finally launch a YouTube channel and leverage the time invested in writing scripts (~10 hours per week) by turning them into weekly newsletters as well.

That never happened either. 🫠

I fell down another rabbit hole, learning everything about the email newsletter business. I've written 125 newsletters since, not once turning them into YouTube scripts or shooting, editing, and publishing videos.

Six years later

My physical and digital spaces are cluttered and mentally draining.

On top of that, the idea of relearning my editing workflow and acquiring all the new skills I'll need to achieve the quality of videos I want feels daunting.

I love editing, by the way, but I'm dreading that first edit, which will likely take 8-10 hours for a one-minute Reel. “Ain’t nobody got time for that!”

Looking back, in trying to make content that looked and felt like Peter's, I created complexity that felt exhausting before I'd shot a single frame.

Peter McKinnon

The gear was supposed to make me ready. Instead, I was prepared in theory and paralyzed in practice.

I couldn't get out of my own way. Ideation came to a halt. I feared doing the thing I was obsessed with. The thing I loved began to feel like a job I didn't want.

I had everything I needed, but I couldn't get past the overwhelm I created.

The studio I built to start became the reason I didn't.

And yet.

Today, I've never been hungrier to start filming, get back in the editing chair, and finally get some long-form and short-form videos out.

I started this with a confession. I'll close with a commitment.

This week, I started by clearing my desk, closing unnecessary tabs and windows, and began setting up my production studio. The only thing left to do is the work itself.

The pull to create is finally bigger than the fear of starting.

Command Center

The gear is out. The cameras are ready. The first Reel ships this month.

What's the thing you're genuinely excited about, but keep finding reasons not to start?

Hit reply and share with me.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

🧠 Empathetic Leadership Can Make or Break AI Adoption
Why AI rollouts succeed or fail has almost nothing to do with the tech.

🎨 Instagram updates algorithm to benefit original creators
The algorithm just shifted in favor of creators who actually make their own stuff.

🔥 'This is fine' creator says AI startup stole his art
The creator of one of the internet's most iconic memes is calling out AI theft.

My Favorite Things 💙
What I’m reading, watching, and enjoying

📘 Co-Intelligence: Living and Working with AI by Ethan Mollick

Listened to this on Libby last week and it was incredible.

A 4.5-hour read that's the best foundation I've found for anyone wanting to understand AI, how it works, and how it's shaping our world. Even with how much has changed since it came out in 2023, everything in it still holds up.

Good morning.

I finally set up Claude Cowork this week, the desktop version of Claude that can read your files and run real tasks.

The setup itself was the easy part. The bigger lesson came from how I prepped for it, and what happened when I asked Cowork to audit my workweek for the first time.

Today, let’s talk about:

• The hack hiding in plain sight that most setup guides skip

• The 5-step setup that turns Cowork from chatbot to coworker

• What happened when Cowork audited my own workweek

Let’s get into it.

— Gordon

Speak naturally. Send without fixing.

Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Not rough transcription you have to clean up. Actual polished text — ready for email, Slack, or any app.

Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.

89% of messages sent with zero edits. Millions of professionals use Flow daily, including teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.

Start flowing free

🤝 How I Set Up Claude As My AI Coworker

I had Claude help me set up Claude Cowork.

The setup guides don't tell you to do that.

The desktop version of Claude can read your files, run real tasks, and act like an actual coworker instead of just a chatbot.

Before I did anything, I asked Claude to write a detailed setup guide for me. 1 hour and 20 minutes later, I had a working AI coworker built around my business.

The hack hiding in plain sight

Setup guides for Cowork are everywhere on Twitter and LinkedIn, but they're all general. None of them are specific to your business or industry.

My wife says it best. "The best hack in the world is to ask AI how you should ask it how to do something."

I had Claude build a custom setup guide for me. It pulled from everything I'd already documented about my business.

• ICP (ideal customer profile)

• Newsletter system

• Positioning doc

• Full tech stack

• Business goals

• Style rules

By the time I sat down to set up Cowork, I had a personalized walkthrough that knew exactly where I wanted to land.

Your AI coworker is only as good as the context you give it.

What I almost did instead

Claude Connectors

Open Cowork without prep, and you'll do what I almost did. Start clicking. Connect every tool the app suggests. Install extensions before knowing what they do.

I know because that's what I did first. Before I'd written a single foundational file, I'd already connected:

• Visual Studio

• GitHub

• Notion

• Zapier

• vidIQ

• Canva

• Figma

I also installed nearly every available Anthropic extension.

Connecting tools is fine on its own. The trouble starts when you hook AI up to your business before you've told it anything about it.

I'm pretty sure I told you not to do that in the newsletter two weeks ago, yet here I went. 🫠

Let me show you how I should have done it from the start.

HERE'S HOW:

• Build a folder structure

• Write four ABOUT ME files

• Set global instructions

• Drop reference files into a library

• Run one real task

Each step earns the next. Skip one, and the rest get weaker.

Build a folder structure

Cowork needs to know where to look and where to leave things alone.

I created a parent folder called CLAUDE COWORK with 4 subfolders, each with a distinct role.

ABOUT ME holds the files Cowork reads before every task. LIBRARY holds bigger reference docs Cowork only reads when I summon them.

OUTPUTS is where finished work lands, and TEMPLATES is reusable scaffolding.

The architecture matters. Lean files for auto-read, deep references for on-demand.

Write four ABOUT ME files

My initial prompt to begin setup

These are the files Cowork reads before every task, so every line has to pull its weight.

4 files in total:

• Who I am and how I work

• My voice and style rules

• My current goals and focus

• Who I serve and what I solve for them

I had Cowork draft them using the source material I'd already prepared.

Total size stays under 6,000 tokens combined. Lean is the goal.

Set global instructions

Added Global Instructions to Cowork

Global Instructions are the rules Cowork follows on every task, no matter what you ask.

Mine looks like this:

• Read every file in ABOUT ME before any task

• Never touch LIBRARY or OUTPUTS unless I say so

• Ask before assuming

• Never fabricate

The folders are the rules. The instructions enforce them.

Drop reference files into a library

LIBRARY is for the docs Cowork should know exist but shouldn't read every time.

If Cowork needs it on every task, it goes in ABOUT ME. If it needs it occasionally, it goes in LIBRARY.

My full tech stack inventory lives there, along with a prompt creator prompt doc.

Cowork knows these files exist. It suggests when to load them. I confirm before it does.

Run one real task

My tech stack audit

The setup only counts when Cowork actually does something for you.

Pick a task you know well enough to spot when the output is off.

I asked Cowork to read my tech stack and identify where my workweek leaks time. Mid-task, it caught a conflict between 2 of my files about how I use Notion.

Instead of guessing, it asked me to clarify. The whole task took 25 minutes.

That was the moment I knew the setup was solid.

What the audit found

Cowork found 4 places where my workweek leaks time. All 4 I knew about. 1 it made impossible to ignore.

I still don't have a clear idea capture system in place. That was the one that stood out.

But Cowork also got one wrong. One of its findings was based on a calendar block that exists in my plans, not in my actual week.

I gave it aspirational context, so it gave me aspirational answers.

AI is only as smart as what you feed it.

In under 2 hours of setup and one test task, I built a working AI coworker and the foundation for everything that comes next.

Cowork only worked because the foundation was already there. A positioning doc written in January. Voice rules refined across 124 newsletters. An ICP crafted to sharpen who I was talking to in my content.

If you don't have those yet, your first ABOUT ME file is where they start.

That's the work most people skip. Then they wonder why their AI gives them generic answers.

Want to set this up for yourself?

Reply with "send the prompts," and I'll personally email you three:

• A prompt to define your ICP

• A prompt to build your positioning doc and topic tree

• A prompt to create custom Cowork setup instructions for your business

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

📊 How to Move from AI Experimentation to AI Transformation
Most companies have adopted AI. Few are seeing the bottom-line impact.

📱 Instagram Improves Insights UI, Adds New Metrics
New tabs, share and skip rate data, and views over time. The metrics most creators have been asking for finally surfaced, where you can actually find them.

🔗 What Technology Convergence Looks Like in Practice
Paradigm shifts come from combining technologies, and AI is the connective tissue.

My Favorite Things 💙
What I’m reading, watching, and enjoying

📖 Unshakeable by Tony Robbins

I read this last week. The insights from some of the world's greatest investors are genuinely useful, and the approach to investing your hard-earned money is simple and thoughtful.

That said, the book felt more like a sales pitch for a few lead magnets, which was a bit of a turn-off.

If you're new to investing or unsure what to do with money that's sitting in savings being eroded by inflation, this is a good book to read (or listen to). At minimum, it'll give you a solid baseline.

Good morning.

How the AI Twitter bros say they're using AI, and how the rest of us use it, are two very different things.

Today, I'm walking you through the five-stage framework I use to decide where AI fits in my business and creative process.

Let’s talk about:

• My AI Fit Framework

• How slowing down speeds up AI

• Why I still haven't fully automated anything

Let’s get into it.

— Gordon

🐢 How To Know Where AI Fits

I've saved sixteen hours a week with AI.

The biggest gains came from the weeks I used it the least.

More and more business owners want to bring AI into their work.

They've seen the hype, pulled up ChatGPT, and they're ready to go.

OK, but now what? They don't know where to start.

Here’s the thing. When you start in the wrong place, AI makes your work worse.

You might want to automate tasks that require your judgment, while AI sits idle on work it could've handled for you in seconds.

The result is sloppy outputs, broken workflows, and a nagging suspicion that the hype is an empty promise.

The fastest way to use AI is to slow down.

Entrepreneurs move fast. They don’t overthink, they just go.

So for AI, they pick a tool, point it at a task, and hope for the best.

Unfortunately, that approach produces the worst version of AI, because AI only works as well as the thinking you give it.

A better prompt won't fix it. A better process will.

HERE'S HOW:

1. Survey Where AI Could Help

2. Run The Workflow Manually First

3. Bring In AI For The Heavy Lifting

4. Keep Your Thinking In Your Hands

5. Only Automate Once You've Earned It

I call this the AI Fit Framework.

The order matters, so walk every workflow through these five stages, step by step.

Step 1: Survey Where AI Could Help

Before you do anything, ask one question. How could AI help here?

You're not committing to anything. You're taking inventory.

Start with a blank page and list every recurring task you handle. Then scan the last two weeks of your calendar for the tasks that repeat and the hours you've spent on them. Mark the ones that drain the most time.

Look for the heavy lifting, the repetitive formatting, and the research that would take you hours. That's where the leverage lives.

If you can fully document a process, AI can do it.

This stage is reconnaissance.

Scout the terrain now so you know what to watch for later.

Step 2: Run The Workflow Manually First

Do the work yourself. All of it.

This is the step everyone wants to skip.

It feels slow, and you'll tell yourself AI could do it faster.

Running the workflow manually is how you find the bottlenecks that waste your time. It's how you see the gaps in your own process, the ones you can't spot from the outside. And it's how you learn what good output actually looks like, so you have a standard to hold AI accountable to later.

Skip this step, and you're teaching AI to do something you've never seen in motion.

That’s when the wheels come off.

Step 3: Bring In AI For The Heavy Lifting

My custom-built prompts and files for Onward Everyday

Once you've run the workflow enough times to know it cold, bring AI in for the tasks that take you the longest and yield the lowest return on your time.

What AI handles well once you've done the reps:

• Deep research and competitive analysis

• Copyediting for voice, active voice, and reading level

• Reformatting your raw ideas into polished deliverables

• Repurposing one idea across multiple formats for social

My newsletter production is the proof. Twenty hours a week dropped to four or six once I built a system in which AI handles the heaviest parts while I direct the work.

That number fell again later. I'll come back to how.

Step 4: Keep Your Thinking In Your Hands

Flash, Flash, hundred yard dash

Here's where I made my biggest mistake.

Early on, I rushed. I gave AI a quality prompt, but let it figure out the rest.

The output was far worse than anything I would've written myself.

When I slowed down and gave AI my full thinking—a paragraph, a brain dump, and the context in my head—the output sharpened, and my workflow sped up.

My newsletter production had already dropped from 20 hours a week to 4-6.

When I stopped rushing and started leading with my thinking, it dropped again to 3-4 hours. I cut another hour or two off my week by doing more of the thinking.

Slowing down sped me up.

The same is true of any human assistant. Without clear input and context, the output won't be worth using.

AI is an assistant. It only performs as well as the input you give it.

Lead with your thinking. Let AI pressure-test it.

Step 5: Only Automate Once You've Earned It

Re-engagement automation in beehiiv

The final stage is full automation, and it has to be earned.

For something to run without you, you have to understand it completely.

You have to have run it manually enough times to document the steps, the decisions, and the edge cases.

Otherwise, you're automating something broken. A broken process at scale is just a bigger mess. That's where you lose trust.

I've been using AI in my newsletter workflow for over a year. The only things I've fully automated are my beehiiv subscriber flows and list cleanup, work that runs the same way every time and doesn't need my judgment.

Everything else stays AI-assisted, because I haven't earned the right to remove myself from more of the process yet.

Plus, it has to be me, me, me. That Same Ol’ G.

AI can do almost anything now.

That's exactly why it matters where you put it.

Survey the terrain before you build. Run the workflow yourself until you know it cold. Bring AI in for the heavy lifting. Lead with your own thinking. Automate only what you've earned the right to.

The tortoise wins this race, too.

Slow is how you go fast.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

👔 When the CEO Becomes the Brand
What happens when the founder's politics and the company's brand become impossible to separate.

📲 Instagram Expands Your Algorithm Tool to Explore
More algorithm control across Explore and Reels. The twist is whether anyone will actually use it.

🔓 Anthropic's Mythos Model Accessed by Unauthorized Users
A small group gained access to Anthropic's most locked-down AI model the same day it was announced.

My Favorite Things 💙
What I’m reading, watching, and enjoying

📖 The Infinite Game by Simon Sinek

I finally got around to this one last week (audiobook format), and it's one of the best books I've read in a long time. Sinek's breakdown of how capitalism has shifted over the last fifty years and what that's done to modern businesses is incredibly timely.

It reinforced something I already believed: more companies should stay private and focus on their just cause, instead of burning themselves out playing a finite game they can't actually win.

Good morning.

I've been excited about AI's new connectors but reluctant to give them the keys to my external accounts, like Gmail, Drive, Calendar, and Notion. I went deep on privacy policies to figure out what's actually going on behind the scenes.

Today, let’s talk about:

• The toggle that determines if your data is kept for 30 days or 5 years

• What happens when you connect your external accounts to AI products

• The three things I'm doing before I connect anything

Let’s get into it.

— Gordon

Protect online privacy from the very first click

Your digital footprint starts before you can even walk.

In today’s data economy, “free” inboxes from Google and Microsoft, like Gmail and Outlook, are funded by data collection. Emails can be analyzed to personalize ads, train algorithms, and build long-term behavioral profiles to sell to third-party data brokers.

From family updates, school registrations, medical reports, to financial service emails, social media accounts, job applications, a digital identity can take shape long before someone understands what privacy means.

Privacy shouldn’t begin when you’re old enough to manage your settings. It should be the default from the start.

Proton Mail takes a different approach: no ads, no tracking, no data profiling — just private communication by default. Because the next generation deserves technology that protects them, not profiles them.

Get free private email

👀 What Nobody Tells You About AI Privacy

For the past month, I've been dying to go deeper with AI.

The new connectors, the agent tools, the potential.

All of it is exciting.

But every time I got close to linking an account, I stopped. It wasn't clear what the downsides might be.

Every tech creator on YouTube is showing you what these tools can do.

Nobody is talking about what's actually happening behind the scenes when you do.

Or about the risks.

I invested the time in reading privacy policies, help docs, and security reports so you don't have to.

The click everyone's racing toward

Watch any YouTube video about AI right now, and you'll see the same thing.

Connect Gmail. Connect Drive. Connect your calendar.

Then watch the magic happen.

The creators are showing you what AI can do with your data.

Nobody is showing you what happens to your data once AI and the company behind it have it.

I use Claude, so that's where I dug in. The privacy questions are the same no matter which tool you use.

What I found in the fine print

When you give an AI tool access to an account, your information passes through the company's servers.

It's stored alongside your chat and retained based on your settings.

Think of AI like a contractor with a key to your house. They come in when you call them, they see what they need to see, and they leave.

But whether they keep notes and share them with the rest of the company depends entirely on your settings.

One thing that surprised me.

Claude only accesses your account information when you specifically ask it to. It's not passively reading your entire inbox in the background.

You ask a question, it pulls what it needs, and that's it.

If I ask Claude to summarize today's meetings, it pulls my calendar events for today. Not my entire calendar history.

Cool, but retention is the most important piece.

Retention decides how long AI companies keep your conversations and what they do with them.

In Claude, there's a toggle called "Help Improve Claude."

Toggle these OFF

With it on, your conversations can persist for up to 5 years and be used to train future versions of the AI. With it off, retention drops to 30 days and is not used for training.

One toggle. Massive difference.

There's an important distinction here. Anthropic doesn't use raw account information, such as your emails or calendar events, to train the AI. They've been specifically excluded for privacy reasons.

But anything you type about that information is treated differently.

Say Claude reads your client's email. That email is protected.

But when you type "draft a reply to Sarah's concerns about pricing," what you typed could be used to teach the AI, depending on your settings.

And once the AI learns from it, that can't be undone. That's what makes the training toggle worth checking first.

I've had mine off since day one.

I wanted to understand what I was opting into before I opted in.

There's also a location metadata setting that shares where you are when you use Claude. I turned that off, too.

It's a small thing, but it's information I don't need to share.

The parachute problem

“And we have three good main chutes out the window.”

Entrepreneurs jump first and build the parachute on the way down.

It's the instinct that makes us good at starting things.

The problem is that when it comes to data and privacy, that approach is dangerous.

For example, by default, Anthropic employees can't read your conversations.

I would expect that.

However, there are two exceptions.

When you submit feedback via the thumbs-up or thumbs-down buttons, the conversation is shared with Anthropic employees. And if Anthropic's automated safety systems ever flag something, a Trust & Safety reviewer could see it.

That includes whatever email or calendar content was pulled into the conversation.

It’s important to know that bad actors can hide instructions inside documents or emails that trick AI into doing things you didn't ask for. This is called a prompt injection. Yikes.

And as we’ve hopefully learned by now, no company's security is perfect. Even Anthropic accidentally made some of its own internal data publicly accessible a couple of weeks ago.

These aren't reasons to avoid AI. There are reasons to use it with your eyes open.

The privacy toggle you ignore in your AI platform decides what happens to your data for the next five years.

What changed for me

Here's the part I didn't expect.

The research didn't make me more cautious. It made me more confident.

Once I understood the risks, I was comfortable moving forward with intention.

You don't need to become a privacy expert. You just need to do three things before you link anything.

• Check your privacy settings (the training toggle and location metadata that I covered above)

• Before linking any account, ask yourself one question: Do I have a real use case for this, or am I just clicking a button because I can?

• If you can't name a specific workflow, don't link it yet

That last one changed my approach. I'm linking my Calendar, Drive, and Notion to Claude because I have specific workflows for each.

I'm not connecting Gmail because I don't have a strong enough reason yet. I'd love to have Claude triage my inbox every morning, but that's a want, not a need.

And I'm skipping the Chrome extension because it burns through tokens too quickly and is painfully slow. Using web search in Claude is faster and more effective.

Not everything needs to be connected and used. Just the things that earn it.

I'm glad I paused over the past month.

The answers didn't scare me off.

They gave me the clarity to move forward with confidence.

This week, I'm finally linking those accounts. Eyes open, settings checked, and use cases clear.

Have you checked your AI privacy settings yet?

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

🤖 How Gen AI Robots Are Reshaping Services
Waymo has completed over 20 million trips. What happens when AI goes beyond chatbots and starts physically showing up in the real world? If you think AI agents on your laptop are a big shift, wait until you see what's coming next.

📸 Instagram Chief Debunks Popular Engagement Hack
Adam Mosseri finally addressed whether reposting your posts to Stories actually boosts reach. If you've been doing this religiously, you'll want to hear what he said. He also dropped a hint about a feature creators have been begging for.

📖 A Simple Guide to Common AI Terms
A plain-language glossary of AI terms from AGI to hallucinations. If you've ever felt lost when someone throws around "LLM" or "chain of thought," bookmark this one. It's the cheat sheet I wish I'd had two years ago.

My Favorite Things 💙
What I’m reading, watching, and enjoying

🎬 Learn 80% of Claude Cowork in Under 20 Minutes by Jeff Su

Jeff is one of my favorite AI creators, and this video is a great example of why I appreciate his work. He breaks down the seven core capabilities of Claude Cowork with real examples and walks you through the setup from scratch. If you've heard of Cowork but aren't sure why you need it, this is the video to check out. It's the reason I'm confident enough to start using it myself this week.

Good morning.

I’m flying home today from a much-needed spring break with family in Cancún. 🏖️

Recharged and ready to get back to work. 👨🏻‍💻

In the meantime, let’s talk about:

• The 90-day stretch that changed AI

• Why I ignored most of it (on purpose)

• What this means for your business right now

Let’s get into it.

— Gordon

🤖 How AI Became More Hands-On

AI didn't just get smarter over the last 90 days. It got hands.

Well, kind of.

The past two months, I've been writing about Instagram strategy. Carousels, ManyChat automations, and Reels.

Meanwhile, every day, my Twitter feed was going bananas.

OpenClaw, Cowork, Perplexity Computer, Dispatch, Computer Use, Meta acquiring Manus, Nvidia building NemoClaw.

One release after another, nearly every day for weeks.

And with every one of them, someone was losing their f-ing mind over it. 🤯

The weight of all of them landing at once is what got me.

That feeling of watching the ground shift while you're busy building sucks.

And all the noise makes it impossible to know what actually matters.

The pattern underneath the noise

The shift happened in a pattern.

OpenClaw went viral because it didn't answer questions. It opened apps, booked flights, and managed calendars.

Anthropic shipped Cowork and Computer Use.

Perplexity launched Computer, a tool that builds and executes entire workflows across 19 different AI models.

Meta bought Manus for $2 billion and started turning WhatsApp Business into an "Action App."

Every major company made the same bet within 90 days.

When the chip company that powers all of AI (Nvidia) announces its own agent platform, you know the era of "ask a question, get an answer" is wrapping up.

That tells you everything.

AI stopped being a search engine on steroids. It became a coworker.

The entire industry pivoted from building chatbots to building agents.

AI that acts.

It opens your apps, fills your spreadsheets, drafts emails and texts in your accounts for your review before you hit send.

It manages your files and builds systems with code while you're away from your desk.

Perplexity claimed their Computer tool replaced a $225K marketing stack in a single weekend. One tool. One weekend.

VentureBeat called it the beginning of the end of the ChatGPT era. Fortune Business Insights projects the agentic AI market will grow from $9 billion this year to $139 billion by 2034.

What I did with the noise

I ignored most of it. On purpose.

OpenClaw required Terminal installation and had real security concerns. Researchers found data exfiltration vulnerabilities. Users reported agents accidentally deleting entire email inboxes during automated cleanup.

I passed.

Instead, I watched where the dust settled. And it settled on Claude.

Cowork gives you a persistent AI workspace. Dispatch lets you assign tasks from your phone.

Computer Use lets Claude navigate your browser, control your desktop, and complete tasks across your apps. All inside one ecosystem, without needing to touch a command line (the text-only interface developers use to talk to computers).

That's when I made a decision. Once I wrap up my current consulting projects and get back from spring break, I'll go deep.

Anthropic Academy, Cowork, Code, Skills, Dispatch. Learn it all in a safe environment before helping others build with it.

I haven't mastered any of it yet. But I know where I'm going.

What this means for you

You don't need to understand every release from the last 30 days. You need to understand one thing.

AI is quickly becoming a tool that does work alongside you.

If you're running a service business solo or with a small team, the gap between you and the company with 10 employees just got a lot narrower. Here's where to start.

• Explore what's new in Claude. Cowork, Connectors, Memory. Open it up and poke around. Don't try to master it yet. Just see what's there.

• Pick one repetitive task in your business and ask yourself whether you'd want an AI to handle it end-to-end. Complete the work, not just draft a response.

• Pay attention to the pace. Since Anthropic launched Cowork in January, they've added Computer Use, Dispatch, Memory, Excel and PowerPoint integration, and a connectors marketplace. All in about 3 months. The features launching today will feel basic by the end of the year.

Remember, anything you can document, AI can complete for you.

Over the past couple of months, AI has been learning to do more than just answer questions.

Now I'm catching up. Intentionally, not frantically. Learning what these tools can actually do before I translate for you.

That's what I love doing. Filter the noise, find what works, and translate it into something you can use today.

I'll be sharing what I find.

What's the AI development that caught your eye recently?

Hit reply and tell me. I want to know what's on your radar.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

🧠 Tapping into Your Team's Circadian Rhythms
Your best work might be trapped in the wrong time slot. This explains why.

🐝 Beehiiv Expands into Podcasting, Taking Aim at Patreon
My newsletter platform* just added native podcast hosting. And creators keep 100% of revenue.

🤖 Perplexity's New Computer Is Another Bet That Users Need Many AI Models
One of the tools I mentioned above is explained. Worth reading if the agent shift caught your attention.

Good morning.

Today is the 52nd consecutive week of publishing Onward Everyday. 🥳

120 newsletters total across 34 months. Not too bad, buddy.

This issue will be a little meta. Not Meta, but meta.

Thanks for sharing this journey with me. 🤝

Today, let’s talk about:

• What a full year of showing up every week taught me

• The machine behind the streak

• The lesson I didn't expect

Let’s get into it.

— Gordon

🎯 What 52 Weeks of Consistency Taught Me

The goal wasn’t 52 weeks straight.

Or perhaps it was.

In April of 2025, I was behind.

I'd started this newsletter in May of 2023 with a single commitment.

Publish 100 issues in two years.

By the time spring of last year rolled around, I was running out of runway to hit 100 without going every single week for the rest of the year.

So I made a simple, practical decision. Not a grand declaration. Just math.

I committed to publishing every Thursday.

What I didn't realize was that the commitment would teach me more about building something that lasts than hitting the number ever would.

The number changed. The method stayed.

When I published my 100th newsletter on November 13th, the goal I'd chased for two and a half years was done.

Then I looked at the calendar and realized I was already deep into something else.

I'd been publishing every single week since April. I was closer to a full year without missing a Thursday than I was to quitting.

So I set a new goal. Not a bigger number, but a different kind of proof.

52 consecutive weeks. One full year.

When you commit to something, the journey won’t be what you expected.

If you keep going, you’ll still get there.

The holiday I didn't take.

For the first couple of years after I started, I'd take a break from Thanksgiving through New Year's Day.

This year, I didn't. I published straight through the holiday season, including the weeks I absolutely did not feel like it.

Not because you needed my newsletter over the holidays.

I’m certain you didn’t.

Because consistency doesn't negotiate with the calendar.

That was the moment the streak became real.

The machine that made it possible.

I didn't publish 52 weeks because I'm more disciplined than you.

I published consistently because I invested time in building a system that made it nearly impossible to fail.

When I started, a single newsletter took me close to 20 hours.

Ideation, research, drafting, revising, and editing. All of it was done manually.

By the time I hit my 100th newsletter, the same process was taking 4 to 6 hours, start to finish.

I built a Claude Project with 7 custom prompts. One for each phase of my production process.

The AI handled the time-intensive parts. Every decision still ran through me.

One of those prompts runs a full editorial review.

It checks for things like:

• Style violations

• Repeated phrases

• Comprehension loss

• Whether the closing circles back to the opening

I wanted to spend my time thinking through the idea and angle, writing the draft, and doing the final edit. Claude could put its senior editor hat on for the copyedit and my custom-built quality assurance process.

The system didn't replace my thinking. It gave my thinking focus.

The system did what motivation couldn't.

The lesson I didn't expect.

🌳 My topic tree

I am not an ideas person.

What I'm good at is building systems that fill gaps in existing structures.

So when I realized running out of ideas was most likely to break my streak, I treated it like any other operational problem.

I built a simple idea-capture system.

Not novel, I know, but a critical system I hadn’t built for myself yet.

I started logging topics in real time.

That meant conversations, questions I kept getting asked, and things I noticed in my work with clients. Then I used AI to organize them into a topic tree. I grouped them by theme, so I always have buckets (branches) to pull from.

The weeks I felt most stuck were almost always the weeks I hadn't been capturing ideas.

Log it now, write it later.

What I'm carrying into week 53 isn't pride in the number. It's what I proved to myself by showing up consistently.

You don't need to be an ideas person. You don't need endless motivation.

You need a clear topic you can talk about for years, a simple template that delivers consistent value, and a system that handles the parts you're not built for.

Build the system. The consistency follows.

Hit reply and tell me the one thing standing between you and showing up every week.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

📖 Don't Let AI Destroy the Skills That Make Your Company Competitive
The hidden cost of AI that nobody's talking about, and it has nothing to do with jobs.

🔄 Instagram Now Lets You Rearrange Carousel Posts After Publishing
One of Instagram's most-requested features is finally here, and it changes how you think about posting carousels.

⚖️ Jury Finds Meta and YouTube Liable for Social Media Addiction
A landmark verdict that could reshape how platforms are built and what accountability for Big Tech finally looks like.

Good morning.

We covered Comment-to-DM with ManyChat a few issues back. This week, we're going deeper into the additional automations that will help convert leads.

We’re talking about:

• Why one ManyChat automation is never enough

• The three flows that cover every high-intent moment in your follower journey

• How to turn warm DM conversations into booked calls on autopilot

Let’s get into it.

— Gordon

Your AI tools are only as good as your prompts.

Most people type short, lazy prompts because writing detailed ones takes forever. The result? Generic outputs.

Wispr Flow lets you speak your prompts instead of typing them. Talk through your thinking naturally - include context, constraints, examples - and Flow gives you clean text ready to paste. No filler words. No cleanup.

Works inside ChatGPT, Claude, Cursor, Windsurf, and every other AI tool you use. System-level integration means zero setup.

Millions of users worldwide. Teams at OpenAI, Vercel, and Clay use Flow daily. Now available on Mac, Windows, iPhone, and Android - free and unlimited on Android during launch.

Try Flow free →

⚙️ What 3 ManyChat Automations Actually Convert

ManyChat is built to automate Instagram conversations.

Most solopreneurs use it for one flow and stop.

You set up the Comment-to-DM automation, watched it work, and felt the pull to keep going. Then you hit a wall.

The options inside ManyChat are abundant.

Triggers, flows, sequences, AI steps, and integrations.

When everything is available, it’s hard to prioritize and know what to build.

Every day, when the next automation isn't running, leads are slipping through the cracks. Someone replies to your Story, but the conversation dies on the vine.

A warm follower DMs a question, gets no response, and moves on.

The foundation is in place. The house is still unfinished.

The three automations that actually move the needle aren't the flashiest ones. They're the ones that work 24/7 without you.

After Comment-to-DM, assuming that the one flow you have is enough is the most common mistake. It isn't.

The goal isn't to automate everything. It's to focus on the three highest-leverage moments in your follower journey. That’s speed-to-lead.

One captures high-intent Story replies before the 24-hour window closes.

Another turns your existing content into a permanent lead engine.

The third converts warm DM conversations into booked calls while you sleep.

The right automations address all three.

HERE'S HOW:

• Story Reply Capture

• Keyword Resource Library

• Lead Qualification + Booking Flow

Build them in the order listed.

The first two generate leads. The third one converts them.

Each takes a few minutes to set up.

Story Reply Capture

ManyChat Instagram Story Reply Trigger

A Story reply is the highest-intent signal on Instagram.

Someone paused, typed, and hit send.

Most creators let these conversations die in the inbox.

The reply comes in and gets buried.

By the time there's a free moment to respond, the window has closed, and the momentum is gone.

Instagram limits businesses to a 24-hour messaging window after any user interaction. ManyChat fires an automated DM within that window, acknowledges the reply, and delivers something useful.

The follower feels seen. The lead stays warm.

The key is designing Stories that invite replies from the start.

A strong opinion works. So does a direct question, or a peek at something you're building.

Think of it this way:

• A bold take gets people to respond

• A generic update gets a tap and scroll

The more specific the prompt, the more qualified the reply. And the easier it is to follow up with something that actually moves things forward.

Set it up once, and every Story reply becomes a doorway instead of a dead end.

Keyword Resource Library

Keywords Trigger in Manychat

This is the automation that generates leads from content you've already created.

Choose keywords tied to your best resources.

For me, it would be:

• CHECKLIST

• PLAYBOOK

• TEMPLATE

• SYSTEM

• GUIDE

When someone DMs any of those keywords, they get the resource instantly.

A follow-up message asks them to share their email.

Then point to those keywords in your bio, your Story highlights, and the captions of your strongest posts. Someone finds an old Reel, sees "DM me CHECKLIST," and sends it at 11 PM on a Tuesday.

The flow fires automatically.

No new post required, no manual reply needed.

The library compounds quietly over time.

Every new piece of content is a potential trigger. Every post you've published is a promotion vehicle for a flow you set up today.

The content doesn't have to be new to be useful. It just has to be pointed at a live keyword.

The content you've already created could have generated leads.

You just hadn't given it a destination yet.

Lead Qualification + Booking Flow

Qualify + Book Leads in ManyChat

The first two automations fill the top of the funnel.

This one converts it.

Once someone has engaged through a Story reply or keyword flow, the hardest part is already done.

The booking flow takes over from there.

The sequence starts with three qualifying questions inside the DM:

• The first surfaces context

• The second reveals urgency

• The third shows whether they're a fit before you commit to a call

Once answered, a Calendly link is added directly to the conversation. ManyChat connects directly to Calendly.

No redirect to a landing page, no extra friction, no manual follow-up required.

The prospect answers the questions, clicks the link, and books.

All inside Instagram.

For service business owners, this is the automation that pays for itself.

A booked discovery call from a DM is a prospect who raised their hand, answered your questions, and put time on your calendar on their own.

This is the automation that turns ManyChat from a content tool into a client pipeline.

Story Reply Capture catches high-intent leads before the window closes.

The Keyword Library turns content into leads.

The Booking Flow turns those leads into calls on your calendar.

Together, they don't just save time. They build a system that works whether you posted today or not.

That's the difference between a tool and a system.

See you next week.

ONWARD TOGETHER.

Everyday Insights 👀
A few things worth reading this week

📖 How Senior Leaders Can Build Their Influence
A strong argument isn't always enough to get people moving. This HBR piece is worth a read if you're working to get peers and decision-makers on board.

🎥 Human Speech and Presence Help Drive Reels Engagement
If you've been wondering whether showing up on camera actually matters, this one has your answer.

🔒 Meta Ends End-to-End Encryption on Instagram
Instagram DMs are becoming less private—and if you're building automations on the platform, this is worth understanding. Learn why Meta made the move, and what it could mean for how your data is used going forward.

My Favorite Things 💙
What I’m reading, watching, and enjoying

📺 The Most Terrifying and Hopeful AI Conversation I've Watched

I watched this one over two sittings.

The first half is genuinely frightening.

Daniel Priestley makes a compelling case for why we could be headed toward another Great Depression by 2029, driven by the unsustainable cost of AI infrastructure.

The second half finds some light. And the ending surprised me.

Both sad and heartwarming, it left me thinking differently about the legacy we actually leave behind.

One of the better conversations I've come across in a while.