The absolute slowest way to use AI is to type at it all day.

You end up trapped in 'prompt-jail,' babysitting a machine that just hands you more work.

But what if you didn’t have to hold its hand? 
What if your AI actually did your work, automatically, and you could get your life back?

The smartest builders in the world have completely spoonfeeding their AI and how they did it just broke the internet.

It’s a self running system and they call it loops.

So I built a free, open-source tool to help you make loops (AI that works for you). Even if, and especially if, you have no idea what the heck a loop is.

It’s called Loop Kit (scroll to the bottom to get it) and it’s here to free you from your desk.

Sponsored

Your prompts are leaving out 80% of what you're thinking.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

the AI hangover

Let me explain. Over the last few months, I’ve shipped a lot. Dozens of open-source projects, triple that for private projects. A dozen sites. My own SaaS. Free tools. Internal tools. Tools I built with founders. Content. Head spinning?

You get the picture. But by May, I felt trapped.

I was jumping from prompt to prompt across my mac, my phone, anything with an internet connection, just to keep my bot army running. I was blasting through work, but I had become a full-time prisoner. For every task I automated (and I automate A LOT), 5 new ones emerged in its place. Like an AI hydra rearing its ugly head at Captain Matt-Merica.

I call this the AI hangover.

And it’s not just me. I talk to founders all day. And I hear the same story every single day. But I hear two sides of it:

1. “I’m AI-pilled and even I’m drowning in it. It keeps giving me more things to do”

2. “I don’t even use AI that much. But now everyone’s sending me 40 page docs I need to read/review. And when I ask my AI to do things, it needs me to hold its hand all day and it’s constant homework”.

Turns out, the masterminds behind Claude Code and Openclaw, were feeling the same way, but they found a way out.

Twitter tweet

Peter Steinberger, the creator of OpenClaw, posted a single take that did millions of views: you shouldn’t be prompting your AI anymore, you should be building loops that prompt it for you. Boris Cherny, who built Claude Code, said it even simpler: his job isn’t writing code anymore. It’s writing loops, and letting the agent ship the work.

Twitter tweet

What do they know that we don't?

Here's what.

The secret is firing yourself as the middleman.

If you keep trying to play the prompt game on hard mode, you are quietly letting your most valuable asset, your own time, dwindle down to absolute zero.

the day it clicked

The first loop I built was for SEO articles. Tedious yes. But I’ve seen enough AI slop writing to know what I didn’t want.

So I took my SEO kit, a thing I'd already launched and helped a bunch of founders implement for themselves, and spun up a loop version just for our new company site.

It grades its own work hard. Makes its own photos. Checks the page before anything goes live.

To break it down really easy:

• My original SEO kit automated the ‘what SEO task should happen next?’

• My SEO loop asks “Is this actually something I’d be proud to publish under my name”. It’s automating the judgement around my taste, my standards, and my approval process.

The first (non automated) pass I did with the AI and it took me about 70 minutes working casually from my phone.

And here's what I want you to hear. 50 of those 70 minutes wasn't the wiring.

It was teaching.

The same way you'd coach a new hire after their first draft. I told it the tiny things I cared about. The proof I want it to show off. The phrases I never want it to use. My voice. My standards.

The kind you only want to do once. So I made sure it was once.

But now when the loop runs it has a judge that checks every future draft against that bar. And that bar gets tighter over time.

Twitter tweet

Once the job finishes, it kicks off a new one on its own. Nobody presses go. Then it just runs.

And when I turned the loop on for my second batch I had 5 more articles ready to roll.

All reviewed, and ZERO edits needed by me. I approved it one-shot. From my phone. Cuddled on the couch with my wife watching the new Yellowstone sequel. Yup. I can be that guy now.

That's the promise in one picture. The work got made well, while I was doing something else.

Reply LOOP with the job you keep redoing and I'll send you the loop shape for it. Free for the first few.

so can I finally "automate everything"?

Automating repeatable work is easy. I do this all the time.

But if you run a business, you know you are constantly juggling new projects. A new brand. A new content format. A new SaaS feature.

And here’s the truth: you can’t hand off 'new' to an AI blindly. New requires your actual judgment.

This is why self-running loops only work when you build what’s called a Judgment Gate.

You automate the plumbing. That part is easy. But the judgment, the hard part, is where you have to bake yourself into the process.

But here is where things get crazy... Machines don't know what magic feels like. They don’t know what a banger social video ‘feels like’. Not yet. So you have to write down exactly what 'good' looks like, and sharpen it every single round.

Think about it. Your first time doing anything usually sucks. AI's no different. Its first attempt never blows me away. It's kind of just meh.

The loop is how "meh" becomes "yup, ship it" without you sitting there every round. Or even being there to get it started.

Twitter tweet

the bouncer framework

People ask me this all the time, so let me be crystal clear. I’m not scared of AI. I jump on new tech fast. But I am ruthlessly careful about what I actually ship.

Because the more I build, the more I realize one undeniable truth.

If you build a skyscraper on a shaky foundation, it tips.

This is exactly why I Loop Kit is built with something called Checkpoints.

All that is is a brutal, hard bar that the AI has to pass before the work moves on. Think of it like a bouncer. The written standard is the VIP list. The bouncer is the gate. If the AI doesn't meet the standard, it doesn't get in.

People think gates slow you down. They don't. Gates are the exact thing that allows you to build tall.

But here is where the panic sets in... Every single week, I hear the exact same two worries from founders.

• One: Is this bot actually going to sound/think like me?

• Two: How do I let it run wild without it destroying my brand?

These are fair questions. I know, because I used to ask them myself. But after building dozens of these systems, the answer’s simple.

Now, I know the idea of 'Checkpoints' can sound a bit abstract. So let's get completely tactical.

How do you actually design a Checkpoint for your own AI system? Before you build anything, you need to ask yourself this one million-dollar question: What is the absolute worst-case scenario if this specific task goes wrong? 

If you’re automating an email newsletter, your Checkpoint isn't 'Did the AI write 500 words?' Your Checkpoint is 'Does this email sound exactly like the brand's voice without hallucinating fake links?'

If you are automating code, your Checkpoint isn't 'Did it run?' It is 'Did it pass all five of our security tests without throwing an error?' 

Whatever that worst-case scenario is, you write a hard, brutal rule to prevent it. That rule is your Checkpoint. That rule is your bouncer."

the loop that took two hours off my desk

But here is the specific loop that completely changed my week.

Every single day, I have to review a mountain of work. Static ads. Videos. Internal projects. Written copy. Before this, that process stole 2+ hours of my day. I was grinding through items, repeating the exact same opinions, and becoming the ultimate bottleneck for my entire team.

So, I made the one move that actually mattered. I took my personal standards, built them into a Loop, and gave it to my team.

Now, the machine applies my opinion and my bar to the work before I ever even see it. They check the work against my standard, without me having to be in the room. It auto-runs. It learns. It gets sharper every single time.

Here is what happens when you build this.

You stop being the bottleneck. That massive block of wasted time completely disappears. You actually get your day back for the high-leverage things that require a human, instead of doing the exact same manual review over and over.

Twitter tweet

If you want to stop getting dragged into the weeds... Reply 'LOOP' in the comments and tell me the one review task you hate doing most. I will personally reply and tell you exactly how I would build a Loop to automate it for you. Free for the first few.

what Loop Kit actually is

So, what actually is Loop Kit? It is three specific skills, packed into one single install. Built for Claude Code, Codex, or any agent that reads skills.

Now, I originally built this for myself. But after talking to hundreds of founders, I realized we’re all on the exact same journey. We’re all just trying to build our brands and figure out these tools without losing our minds. So, I’m open-sourcing the exact system I needed, and I am giving it to you.

It is broken down into a simple, three-part architecture:

loop-kit/
  loop-start    → the front door. plain questions, runnable system out.
  loop-builder  → the full design when you know what you want.
  loop-doctor   → the safety check. "is this safe to run, and where will it hurt me?"

loop-start

Think of this as the front door.

You don’t need to know architecture jargon. You don’t need to write complex risk codes. It just asks you one or two plain-English questions, and immediately hands you a system you can actually run. You just answer the questions, and it maps the loop.

loop-builder

This is for when you already know exactly what you want.

It maps the entire loop, writes the plan, and gives you a prompt for a no-stakes dry run. It produces the smallest, safest version that actually gets the job done. Map it. Plan it. Dry-run it. Ship it.

loop-doctor

This is the part that actually lets me hang out with my kids at night.

Before any loop gets scheduled, it asks one brutal question: Is this safe to run, and where will it hurt you?

If anything is unclear, it completely stops instead of plowing ahead.

The default is 'don't,' never 'send it.' It’s the ultimate safety net running underneath everything you build."

the four questions I never skip

Under all three skills is one stubborn little check. It's the whole trust layer.

Before I let any loop run, it has to answer four dead-simple questions. The same ones I'd ask a new hire before handing them the keys.

Can't answer all four? Then it's not a loop. And the kit will tell you so, and hand you a checklist instead.

the failure modes it's watching for

This is the part I'm proudest of, because these are the exact ways a good loop quietly goes bad. loop-kit names them so it can catch them.

→ It says "done" when it isn't. A soft "done" that hides a half-finished job. That’s the Ralph Wiggum loop.
→ It keeps running long after nobody on the team understands it anymore, and the unread output piles up. That's comprehension debt.
→ You quietly stop checking the output and just trust it. That's cognitive surrender.
→ You leave it running unattended and it becomes a door you left open. Passwords leak, permissions creep. That's the security tax.

There's one number I watch on every loop too. If the loop keeps handing you changes and you keep saying no, it isn't helping.

So I ask: am I accepting more than half of what it proposes? Under half, the loop is losing. Kill it or fix it.

the old way vs. now

The old way is you, in the chair, every run, forever. Every draft, you. Every review, you. Every handoff between Codex and Claude and whatever's next, carried by hand. You.

Add it up and it's 2+ hours a day on reviews alone for me. Plus the 50 minutes of teaching baked into every new thing I build. Plus the prompt I write to make the prompt to build the system. Forever.

loop-kit gets you the system once and gives you your life back.

You teach it your standard one time. It holds that standard on every run after, while you're somewhere else. The old way costs you the rest of your year, an hour at a time. This costs you one good afternoon of setup.

For free.

get loop-kit

Loop kit is open source. MIT. Free.

You get:

→ three skills: loop-start, loop-builder, loop-doctor
→ the safety engine: the four-question, stops-when-unsure check that every loop has to pass
→ the named failure modes (Ralph Wiggum loop, comprehension debt, cognitive surrender, the security tax) baked into the audit
→ a pattern library index of proven loop shapes, with titles, sources, and links
→ supporting scripts for onboarding, plain-language translation, pattern selection, and a self-audit
→ reference playbooks: archetypes, loop composition, spec and launcher templates, onboarding, red flags, example loop packets
→ install.sh for Claude, Codex, or both, plus doctor.sh
→ tests, a golden-case baseline, MIT license, and a SOURCES notice

No upsell. No catch. MIT license. Fork it.

54% of Americans are sick of AI. They’re experimenting every day, and they’re exhausted.

That’s the AI hangover.

It's the exhaustion of babysitting a problem-machine. It's getting bloated 30-page documents you didn't ask for. It's not trusting the work, and ending up re-reviewing it all yourself anyway.

But here is the undeniable truth... This sickness is just a sign of premature systems. The bet on AI wasn't wrong. My belief in it hasn’t changed.

Because right now, there is a team out there completely reinventing how they work. And they’re going to out-ship and out-sell you a thousand times over.

The entire game is getting the most signal from the noise. You have to build systems and Loops where you get massive output for the absolute least amount of time.

That’s exactly why I built Loop-Kit to cure my own AI hangover. It’s the smartest thing you can do for your bottom line.

Go big,

Matt"

P.S. loop-kit is the on-ramp layer. It's what gets AI running for you safely. Point your loops at the work you've already systematized: Frontrun for market voice, Short-Form Idea Engine for video ideas. loop-kit is the layer underneath that makes those run on repeat without you in the middle.

P.P.S. If you want my team to map your first loop with you, reply "loop" and tell me the job you keep redoing. I'm taking a few a week.

P.P.P.S. Star the repo if you grab it. It tells me to keep building.

In college, I talked WPP out of a million dollars.

Late 2000s. WPP was the biggest ad holding company on the planet, and I was a kid with a pitch: a YouTube competitor, back before YouTube even had an algorithm. The money turned out to be the easy part. Getting anyone to show up was the hard part.

So I sat in the back of class, half-listening, posting to Reddit off an iPod Touch. That’s where I stumbled onto the most unfair traffic move I’ve ever run. I’ve run it for 20 years since, and it just quietly became more valuable than it’s ever been.

The move: stop chasing traffic and go stand in the conversations your buyers are already having. The threads where people ask “which tool should I use?” are the highest-intent traffic on the internet, and nobody owns them. You just show up and be useful. I wasn’t marketing. I was a community member.

Within a year we were doing 500,000 users a month. And Reddit was our engine.

That was almost 20 years ago. I’ve run the same play since, for D2C brands, for B2B, whatever has a community.

And I’ll tell you what I tell every founder who asks me: Reddit is the single hardest platform to crack, for exactly one reason. Show up to promote and you get eaten alive. Show up to be useful and you win. Participation and value.

That’s the whole game. It was true in 2007. It’s more true now.

It finds the conversations your buyers are already having on Reddit, hands you a genuinely useful comment for each one, and keeps you showing up on a steady rhythm without torching your account. Clone it today.

But first I want to show you why this works now better than it ever has, because if you skip this part you’ll just get yourself banned.

Sponsored

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

the part that’s almost unfair now

Here’s what I couldn’t have seen coming back then.

Those threads I was leaving comments on don’t just get read by Redditors anymore. They get read by the AI.

Ask ChatGPT, Perplexity, or Google for the best tool in a category, and they don’t pull your homepage. They pull the page that already ranks for that question. For buyer questions, that’s almost always a Reddit thread.

Check this out. A massive new data study analyzed 30 million AI sources this year.

Reddit came out as the single most-cited source on the entire internet, crushing YouTube and LinkedIn.

Google is literally paying Reddit $60 million a year to train on its data. OpenAI is reportedly paying around $70 million. The biggest tech platforms on earth are spending nine figures to put human Reddit conversations inside their answers.

I call this The Answer Wedge.

  buyers ask  ─►  thread ranks  ─►  AI reads it  ─►  AI repeats it
  on reddit       on google         for answers       to the next buyer
                                         │
                                ┌────────┴────────┐
                                │  your answer    │  ◄── the wedge
                                │  goes in here   │
                                └─────────────────┘

stop overthinking geo

You’re hearing a lot about GEO and “AI visibility” right now. Like everything: mostly noise, a little signal, a few real operators.

Here’s my read: GEO is SEO. Same job, new surface. You’re still optimizing your brand for how people search and discover, the search box just turned into a chat. Do SEO well and your site, crawlability, and authority are already handled. (If they’re not, reply and I’ll send you my one-pager.)

Check out what Google said about it.

What’s actually new: your website is now just one piece. Reddit, LinkedIn, and YouTube do the heavy lifting in AI answers, and Reddit is the one founders ask me about every single week.

They know they should be there. They’re just intimidated.

• Where do I start?

• Which subreddits?

• Use my name or stay anonymous?

• How do I not get banned on day one?

That’s exactly what I built this for.

here’s what it actually does

Now, you could do all of this by hand. Spend your hours scouring subreddits, guessing which threads actually matter, writing every response from scratch.

But if you’re like me, you don’t want a second full-time job living inside Reddit. You want the workflow on easy mode.

Want more easy mode? We install custom AI marketing agents that do 100% of the work. Book your build here.

So I built a tool that runs the whole thing for you. You point it at your category. It goes and finds the Reddit threads your buyers are already reading, the ones already ranking on Google, and ranks them by how worth your time they are.

Then it hands you back one clean page. Every thread, sorted into the ones to enter now, the ones to keep an eye on, and the ones to skip. Something like this:

And for each one worth entering, it drafts you a comment you can finish in about two minutes. Not a pitch. A real answer: lead with the honest options, say plainly where you lose to Obsidian and where you win, and disclose who you are. Then you make it sound like you, and post.

That’s the whole loop, on one page, every time it runs.

how much it does for you is your call

Think of it as a dial, from “just show me” to “handle it.”

At the start it just does the homework: here are the threads, here’s a comment for each, you make it yours and hit post. Nothing leaves without you.

Once it’s earned your trust, you let it post the ones that clearly pass your rules, on your schedule, and it taps you the second anything looks off.

the one rule that actually matters

Twenty years of this comes down to one rule, and it’s the opposite of what you’d guess.

The whole time I was pulling in those half a million monthly users, only one in every seven things I posted to Reddit was our content. The other six posts were just useful to the community. Six for them. One for me. Value to promotion.

That ratio is the move. So I built it into the code, where you can’t break it by accident: be valuable, not a scammer.

Spray promotional comments and Reddit will eat you alive, exactly like it would have in 2008. So the kit won’t let you.

→ It paces you. A cap on how much you post a day, how often you hit the same subreddit, and a cooldown in between. It would rather post less than build something that looks like spam.
→ It makes you disclose. A comment that hides who you are doesn’t go out. Period.
→ It pulls the plug the moment things go sideways. One comment removed, one annoyed mod, and it stops everything and tells you. You start it back up on purpose, never by accident.

One real account. Real answers. Honest disclosure. It’s the only version that has ever worked, and the only version that ever will.

wait for it…

The payoff lags. Reddit moves first, Google later, the AI tools later still. Give it 30 to 90 days, not 30 minutes. Anyone promising you a guaranteed ChatGPT citation in two weeks is selling you something.

who this is for

→ founders in categories where buyers ask “best tool for X”
→ agencies building AI-search visibility for clients
→ anyone willing to write one genuinely useful comment before they mention themselves

Not for you if:

→ you want to run bot farms or fake accounts. this breaks on purpose for that.
→ you can’t disclose who you are. the gate won’t let you.

why I actually care about this

About a year ago I got to spend half an hour with Alexis Ohanian. I’m not telling you that to name-drop. I’m telling you because I mostly spent it thanking him.

Reddit didn’t just build my career. Back in 2020, on a few dead-quiet niche subreddits, it’s where I found the other weirdos who were obsessed with AI before it was a thing. I love a good rabbit hole. Reddit is the best one ever made.

This kit is me handing you the on-ramp I had to figure out the hard way.

get the kit

Point it at your category, pick how much you want it doing for you, and start showing up in the answers.

You get:

• the part that finds the Reddit threads already ranking on Google for your buyers

• a score on each one, so you know which to enter and which to skip

• a ready-to-finish comment for every thread worth your time, your disclosure already in it

• the guardrails that keep you from getting banned: pacing, disclosure, and an automatic stop the second a mod pushes back

• the dial from “just show me” to “handle it”

• a weekly read on whether ChatGPT, Perplexity, Gemini, and Claude have started naming you

No upsell. No catch. MIT. Fork it.

The play that worked in 2007 works better today. Go be useful.

Go big,

Matt

P.S. This slots into the stack. Frontrun maps where your category is moving. Slideshow Kit ships the angles. Outcome Kit scores which ones made buyers. The Reddit AI Kit earns you the mention when a buyer asks the AI directly. Four free kits, one loop.

P.P.S. Want my GEO one-pager, the top tips for the rest of the surface (your site, crawlability, authority)? Reply to this email and I’ll send it. And star the repo if you grab it, it tells me to keep building.

The absolute worst thing you can do for your business is guess what your customers actually want.

You end up finding out months and thousands of dollars later that you were completely wrong.

I call this the guessing tax.

But here's the truth... the exact answers you need are already sitting right there. You just can't read 4,000 reviews and a thousand TikToks fast enough to catch market trends in time.

So I got tired of paying that tax, and I built a solution.

It's a completely free, open-source tool called Frontrun (scroll down to clone it).

Frontrun reads your entire category and breaks down two things:
1) where is my category moving, and 2) where are my competitors exposed

And if you're lucky, it can even spot a lottery ticket, a trend that's about to pop before it becomes a bidding war.

Let me show you exactly how it works.

╭─ frontrun · how it runs ─────────────╮
│ url in                               │
│    │                                 │
│    ▼  read the whole category        │
│    │                                 │
│    ▼  stage every angle              │
│    │                                 │
│    ▼  the move for each              │
╰──────────────────────────────────────╯

Sponsored

You Can’t Lead If You’re Stuck in the Middle

Most business owners don’t notice when it happens.

They start as the operator… and slowly become the center of everything.

The real risk isn’t being busy. It’s being required for everything.

BELAY’s new guide From Operator to Owner: How to Exit the Middle Without Losing Control shows you how to step out of day-to-day execution while keeping visibility, quality, and momentum intact.

BELAY matches you with U.S.-based Assistants who bring structure, follow-through, and operational clarity so work keeps moving without you at the center.

Download Operator to Owner for Free.

I scraped 119,411 user conversations in 30 days.

Yup. I scraped 119,411 user conversations in 30 days.

This is for a big project where every convo, every creator, every hook, is tagged, broken down by source, angle, geographic, demographic, format, you name it. The moves we’ve been getting from it are nuts. And while that builds running at serious scale, I pulled one slice out for you.

That’s Frontrun.

And the first thing Frontrun did was prove me wrong.

Let me tell what you happened.

This ingredient kept blowing up my feed. Magnesium L-threonate. I don’t usually get supplement content but I kept seeing creators stacking it.

And the thought crossed my mind… “This thing's about to pop. I could grab a few friends already in the space, spin up my GTM agents, and pull some cash out before the window closes”

So I threw Frontrun at it.

Search demand for threonate was already at 74% of its all-time high. Everybody knew about this thing except me. Ol’ Matt is late to the party.

I wanted a hockey stick. What I got was a flat line that peaked months ago.

And here's the part that matters for you. That same check runs on your own brand, every time you point it at your category. It lays the whole board in front of you. What's still climbing. What your customers are already done with. Where your competitor's exposed. Then you make the call. Five minutes, receipts in hand.

That board is the product. Let me show you one.

(Or skip the setup, reply to this email and I’ll run it for you. Free for the first few

the map you get every single run

I pointed it at magnesium. The whole category, not one idea.

It read the market and handed back the board. Ten angles staged:

Seven angles are rising. Ride those now, while there's still room. Three are mature. People still want them, but the category's crowded, so you win on language, not on novelty/angle. Zero are pre-wave this run. Nothing's early here.

Now I have a map of exactly where every angle stands and what to do about each.

This is what works for your category, whatever you sell. Magnesium, mattresses, B2B payroll software. Point it at the URL, get the board.

Reply to this email with your category. We’ll run it and send you the read (free for the first few)

where your competitor is exposed

Half the map is YOUR angles. The other half is your competition.

Take oxide, one of the three mature magnesium angles. It still pulls heavy search, near the top of the category. So a competitor sees that volume and leads their whole pitch with "high-potency magnesium oxide."

The tool's read on that lane: the category already settled here.

That's the second thing every run gets you: where a competitor is spending cash on angles their customers don’t want.

Frontrun pulls that straight from the reviews and comments, (in the customer's own words) and shows you the opening.

Every brand has a competitor who's exposed somewhere. And only a select few categories find a pre-wave trend.

run it again next week

A map is a snapshot. But the big value is mapping the data over time.

Run Frontrun on the same category next week, and next month, and the angles move. A rising one tips into mature. A mature one starts to die and frees up the budget you were wasting on it. The thing creators were whispering about finally hits search.

One run tells you where you stand today. A standing weekly run tells you where the category is heading before your competitors feel it. That's the difference between reading the market once and watching it move.

Point it at your category. Then keep pointing it.

you can check every line yourself

A map is worthless if it makes things up. So this one can't.

A $5k/month Meltwater sub (or Brandwatch Sprout, etc) just counts. It pulls 4,000 comments and tells you "pricing" was mentioned 47 times. So you slap "address pricing" on a deck, and nothing gets done.

Frontrun ranks by usefulness instead. Same engine on a SaaS pool here, the math doesn't care what you sell:

The 3-mention line beats the 47-mention cluster, 26.0 to 3.2. Mentioned 15 times less, worth 8 times more. The loud, useless theme gets barred.

The specific, commercial one gets promoted.

i ran nearly 300 trends through it. one survived.

So that's the floor. Here's the ceiling.

Every so often, buried in the noise, Frontrun finds the one shape it actually hunts for: creators loud, search still empty. The thing nobody's clocked yet. To see how rare that really is, I ran some tests.

These are six categories printing in ecomm right now:

• Skincare actives.

• Supplement actives.

• The GLP-1 gold rush.

• Functional drinks.

• Pet supplements.

295 candidates, everything hot across all six:

Take skincare. It threw out every ingredient you've already heard of and showed me where each one really sits:

That number is how hard people are searching each one right now, out of 100. The word next to it is the verdict.

Brutal news if you’re looking for novel arbitrage. ALmost everything blowing up your feed is already late. You’d be buying the top.

But one lane out of 295 stayed open.

the one survivor

So what made it through all 295? One ingredient.

Spilanthol. Acmella oleracea. The one creators are calling "nature's Botox", a tingly flower extract they're pitching as Botox without the needle.

Here's why it lived. Search sits at 14.5 out of 100, near the floor. And it's never had a real wave. Most low terms are low because they already peaked and crashed. But this one's low because it hasn't happened yet.

Meanwhile creators are catching on:

There’s even a brand is already shipping a "Fountain of Youth" face oil built on it.

The language is jumping into Spanish. There’s a creator's who’s hook is that a billion-dollar cosmetics giant patented it and buried it.

Creators loud. Search empty. In 295 tries, it found it once.

On September 1 I’m going to run the numbers again (and I’ll post the results). If search has jumped past 20 and the creators are still at it, the call hit. If it's flat and they've gone quiet, it was noise, or we’re still early.

If I'm right, whoever moves on it now owns the lane before it turns into a bidding war.

what it replaces

The old way to learn all this: an $8k research agency, a $5k positioning strategist, three weeks of ad spend confirming what your reviews already said. Call it $35k a quarter, and at least a month gone.

Frontrun gives you the same map in five minutes.

And the real power isn't the one run. It's running it every week and watching the board move, rising terms cooling off, dead ones you can finally cut, a new lane cracking open before anyone else sees it.

Or skip the setup. Reply with your brand URL and we'll run the whole map for you, free for the first few who ask.

who this is for

Point Frontrun at your category if:

→ you want a map of where every part of your market stands, rising, mature, or dead, with the move for each
→ you're sitting on reviews, comments, and competitor ad libraries you've never actually read
→ you want to know where your rivals are exposed before your next campaign, not after
→ you'd rather hear "this isn't early" than get flattered into a bad bet

Not for you if:

→ you've got a research-ops team tracking your category's demand stages in real time. you're the 1%.
→ you want a button that spits out ten confident trend calls from no input. that's the slop machine this exists to beat. mine killed 294 out of 295.

get frontrun

The full engine is open source. Point it at your category and read the map yourself. (You wire up your own scraper API keys. The README walks you through it.)

You get:

• every angle in your category staged rising / mature / dead, with the move for each

• the objection map, where your competitor bleeds, in your customer's words

• the usefulness-ranked angle slate, the 3-mention gold over the 47-mention noise

• the actual ad copy and landing hero, every line traceable to a real quote

• the rare pre-wave call when one's actually there, checked against real search so it's never a mature term in disguise

No upsell. No catch. MIT license. Fork it.

Stop paying the guessing tax. Start reading the map.

If you have been building flows and custom builds.

Go big,

Matt

P.S. Frontrun is the layer that sits in front of the rest of the stack. Slideshow Kit ships the angles to your feeds. Outcome Kit tells you which ones produced buyers. Frontrun maps the move, Slideshow posts it, Outcome scores it.

P.P.S. Frontrun is opensource slice. If you want the full power of the build it came from, reply to this email with your brand or project. That's the whole machine: hundreds of thousands of user and creator voices, sliced by demographic, geography, and segment, delivered as a daily Intelligence Briefs, alerts, and a dashboard that shows you exactly where your market is moving and who's moving it first.

P.P.P.S. Star the repo if you grab it. It tells me to keep building. And in six weeks, hold me to the spilanthol call.

AI was supposed to take work off your plate.

Instead, you got a new job. Babysitting AI.

Babysitting is for 14-year-olds watching the neighbor's kids and eating their snacks. Not founders trying to run a company.

But that's what most AI work feels like right now.

You brief the agent. Paste the context. Correct the assumption. Move the output to another tool. Explain what the last agent did. Decide what should be remembered. Then you do it all again tomorrow.

Twitter tweet

And the problem isn't that AI sucks or can’t do work. The problem is you’re stuck in the middle of the AI dance. You’re stuck in the middle.

So I built a simple system around my agents. It's called Offload Kit. File-first. Agent-native. MIT. Open source.

Many agents in. One human glue layer out.

Brief → Task → Response → Synthesis → Memory Approval → Vault Update

Scroll to the bottom for the link.

Sponsored

A $200M+ DTC brand has 44 people messaging Viktor every day.

Their ops team built inventory command centers and reorder dashboards through Viktor. Supply chain gets daily stockout alerts before they happen. Marketing tracks ROAS and runs content calendars. CS has CSAT scores and support tickets triaged and briefed every morning in Slack, before the first support call. No dashboard digging.

48 internal apps, built through conversation. No code. No developer queue. Command centers, inventory dashboards, sales trackers, reorder systems.

That's one company. Across the platform, teams have built 2,000+ apps the same way: message Viktor in Slack, describe what you need, get a working tool deployed. No code. No six-week dev queue.

Your team doesn't wait for a product roadmap. They message a colleague.

5,700+ teams. SOC 2 certified.

"It was almost instantly adopted by the bulk of my team." — Boris Wexler, CEO, Space Dinosaurs

Start free. $100 in credits →

AI gave you a new job (you didn’t apply for)

The pitch was simple. AI agents would do the work. Research. Write. Code. QA. Analyze. Summarize. Ship.

And sometimes they do.

But zoom out. A lot of founders didn't get a worker. They got a junior employee who forgets the what’s going on every morning.

And I’m not saying that because I see people complain on X.

I talk to founders every day. Some are hiring us to build custom AI flows. Some are consulting with me on how to use these tools inside their business. Some are already neck-deep, using agents all day, trying to turn the messy parts of their company into repeatable workflows.

The pattern is always the same: the people using AI the most are not always doing less work. A lot of them are just drowning in a new job, managing the AI.

So the founder becomes the manager. Every chat starts the same way:

That last one's where the whole thing starts to crack.

With one agent, you can brute force it. Keep the thread alive. Scroll up. Paste another wall of context.

Once you've got multiple agents, that falls apart fast. One agent researches. Another writes. Another reviews. Another builds. Another QAs. Each one needs enough context to do the job without dragging you back into the middle.

If you're still personally carrying every handoff, you don't have an agent system.

You have a group chat with homework.

Twitter tweet

more context isn't the fix

The default answer is "just give the model more context."

That helps. It doesn't solve the actual problem.

Karpathy's been calling this "context engineering," and he's right that the real work is getting the right information into the model before it starts.

Twitter tweet

But long context makes one thread smarter. It doesn't create an operating system for work.

A long chat remembers what happened inside that chat. The useful parts are still trapped unless they get turned into something durable: a project decision, a source note, a handoff, a playbook, a task record, a memory update you can actually inspect.

Hidden memory has the opposite problem. It can remember too much, too quietly.

If an agent silently saves a half-correct assumption as "company memory," every future agent gets worse. Bad summaries become source of truth. Temporary context turns into policy. The system starts sounding confident because it's confidently reusing old sludge.

The fix is better separation.

Raw work goes in one place. Durable knowledge goes somewhere else. You (Mr or Mrs Human) approves what becomes permanent.

my Vault is my secret weapon

I tell people this all the time. My Vault is my secret weapon (I use Obsidian to read it).

It's where my projects, client context, decisions, open loops, source material, playbooks, prompts, examples, and past runs live.

The point isn't that Obsidian is magic fix. It's not. Pick whatever folder structure you want.

The point is that my agents don't need me to recreate the company/project/workflow knowledge from scratch every time they start working.

They read the project context. They see what was decided. They see what's still open. They see the playbook. They see what previous agents tried.

That changes the job. Instead of me saying "here's everything you need to know," the agent starts by reading the notebook.

But the Vault alone isn't enough. The Vault is durable memory. It can't be full of every messy task, half-baked output, and agent scratchpad. The moment that pollutes the Vault, future agents start trusting bad data.

That's where the second piece comes in.

the bus is where work happens

That's the whole system.

The Agent Bus is for work in motion. Tasks. Assignments. Status. Responses. Blockers. Handoffs. Syntheses.

The Agent Vault is for durable knowledge. Project context. Decisions. Open loops. Source indexes. Playbooks. Templates. Approved learnings.

The split is the whole game:

If it answers "what happens next?", it belongs in the bus.

If it answers "what should future agents know?", it belongs in the vault.

This is the mistake that’s easy to make (and I learned this the hard way). You either keep everything in chat (work disappears when the thread dies) or you dump everything into memory (the memory becomes garbage).

You need both. Not because folders are exciting. Folders aren't super exciting I don’t think. But because the split gives agents a way to coordinate without making you the dispatcher.

the five-part offload system

Under the loop are five pieces. Each one removes a specific job you're doing by hand right now.

1. Work Queue        → tells agents what to do next
2. Project Context   → tells agents who the company is
3. Handoff Record    → tells the next agent what just happened
4. Digest Routing    → tells the system where output should live
5. Memory Approval   → tells the agent it can't rewrite canon alone

1. Work Queue

This is your task file. And it’s not a buried chat instruction from 90 minutes ago.

It says who the task is for, what the objective is, what context to read, where to put the response, and what "done" looks like. Boring on purpose. Boring's what stops you from re-explaining the same job to four agents.

2. Project Context

Project context is what the agent reads before starting.

For me, that means the project brief, decisions, open loops, recent run records, and any local agent instructions.

The goal is simple: the agent should arrive briefed.

3. Handoff Record

Every agent should leave receipts.

What did it do? What did it find? What changed? What's blocked? What should the next agent know?

That’s what keeps one agent from making the next agent start cold. No receipt, no handoff, no system.

4. Digest Routing

After a run, useful output needs a place to go.

Some things are tasks. Some are links. Some are project facts. Some are decisions. Some are playbook improvements. Some are just raw notes that should stay in the bus and never become memory.

Digest routing's the rule that keeps your system from turning into one giant junk drawer.

5. Memory Approval

Don't let agents silently rewrite what your company remembers.

If an agent thinks something should become permanent, it proposes the change. You review it. Then it gets accepted, edited, or rejected.

Technically it's just a before-and-after view of the files the agent wants to change. In plain English:

The agent has to show what it wants to remember before it becomes canon.

If you would not let a junior employee silently rewrite company process, don’t let an agent silently rewrite company memory.

what offload actually ships with

This is where the idea becomes useful.

It ships with the pieces I wish every agent workflow had on day one:

offload/
  00-start-here.md
  README.md
  AGENTS.md
  install.sh

  .claude-plugin/
  .codex-plugin/

  skills/
    agent-manager/      → walks the full loop
    new-task/           → creates a queued task
    respond/            → reads task + context, writes response
    synthesize/         → combines responses into one handoff
    propose-memory/     → writes a memory approval packet
    promote/            → applies an approved packet to the vault

  agent-bus/            → where work happens
  vault/                → what the company knows
  docs/                 → first-30-minutes, bus-vs-vault, memory-approval
  examples/             → newsletter planning, SaaS homepage improvement

The skills are intentionally thin. They don't schedule tasks. They don't run a queue server. They don't auto-write permanent memory. They teach the agent how to use the files.

Drop the folder into any workspace your agent can read. Codex, Claude Code, OpenClaw, Hermes, Cursor agents, any file-reading agent picks up the same contract.

Two examples ship in the repo:

The same loop runs for code, ads, design, ops. Any place you've got more than one agent in the chain.

The honest limit: Offload doesn't make agents autonomous by itself. It gives them rails. Agents still need something to tell them a task exists. If you need millisecond coordination between hundreds of workers, use a real queue. This is the starter pattern for people who want the file workflow to work before they build the machinery around it.

this newsletter was planned with the offload kit

This newsletter was planned using the system it describes. That part matters.

I asked Codex to find the angle. It wrote a shared brief into the Agent Bus. Then it assigned separate tasks:

OpenClaw came back with the line I keep using:

That went into the synthesis. Then the synthesis went into the newsletter spine. Then the kit got built and shipped to GitHub. Then Codex QA'd the repo: plugin JSON, install flow, skill files, public links, license, privacy scan, no-em-dash check.

That's the point.

The article was made with the agent coordination system.

Jason Liu at OpenAI had a related point about Codex becoming a broader system for getting computer work done: durable threads, tools, artifacts, queues, browser use, connectors, automation. That shift's why this matters. Once agents move from "help me code" to "help me operate," coordination becomes the product.

Twitter tweet

what offload kit refuses to do

The five rules baked into every stage. Same energy as the rules I put in every other kit, because the same rules apply.

→ Never auto-write memory. Every promotion to the vault passes through a human-reviewed packet. No exceptions.
→ Never trust raw logs as canon. Logs are useful. They're not the company notebook.
→ Never let every agent write to the vault. That's how you get memory pollution. If everything's memory, nothing is.
→ Never make this an "AI memory" project. That framing's too narrow. The real pain is that you're still managing the agents by hand.
→ Never overbuild the bus. You don't need Kubernetes for agents. You need a queue, context, response files, handoffs, and a review step.

If you've ever watched an AI workflow rot because someone let the agent quietly rewrite a doc nobody approved, these rules are the fix.

who this is for

Run Offload Kit if:

→ You use more than one AI agent or chat
→ You keep pasting the same project context into every new thread
→ Agents finish work but the next agent has no idea what happened
→ Useful lessons disappear into transcripts you'll never reread
→ You want memory you can inspect before it becomes permanent
→ You want a starter pattern before building custom automation

Not for you if:

→ You already run one heavyweight agent with deep persistent memory and zero context drift. Keep it
→ You want a SaaS dashboard with a billing meter. That's not this
→ You want hidden, automatic memory the agent rewrites silently. That's the thing Offload exists to prevent

get offload

The first version of Offload Kit is open source. MIT. Free.

It includes:

→ a file-first Agent Bus and Agent Vault
→ 00-start-here.md and a first-30-minutes walkthrough
→ AGENTS.md rules for any file-reading agent
→ task and response templates
→ project context, digest, and memory promotion playbooks
→ the memory approval packet (the diff-before-canon pattern)
→ a newsletter planning example
→ a SaaS homepage improvement example
→ Claude Code plugin metadata
→ Codex plugin metadata
→ six thin SKILL.md workflows
→ install.sh for Claude, Codex, or any agent that reads SKILL.md
→ MIT license

The goal isn't to make you use my exact folders.

The goal is not to give agents infinite memory. The goal is to stop being the memory, router, and handoff layer yourself.

Give them a place to get work. Give them a place to read context. Make them leave receipts. And make them ask before rewriting the company notebook.

That's when agents stop feeling like another tab to manage and start feeling like workers that can actually get shit done.

Go big,

Matt

P.S. Offload is the coordination lane that sits underneath the rest of the kit stack. If you're already running Slideshow Kit, Short-Form Kit, Brand Shoot Kit, Outcome Kit, or any combo, this is the shared filesystem contract that keeps the human from becoming the system every time two of those kits need to talk.

P.P.S. If you build something useful on top of Offload, the kit got out of your way. That's the whole bar.

P.P.P.S. Star the repo if this helps. It tells me to keep building

Your content calendar is blank again.

So you do what everyone does. You open TikTok. You find a competitor whose videos are working. You watch ten of them. You screenshot the three that hit hardest. You make your own version of the best one.

It gets 400 views.

You copied the winner and the copy died. Then you do it again next week, because the calendar is still hungry and copying competitors is the only research process anyone ever taught you.

Here's the brutal truth. You copied the topic. The topic was never the thing that worked.

The thing that worked was a mechanism buried inside the video. A curiosity gap. A counterintuitive claim. A before/after with no narration. That mechanism is transplantable. The topic is not. Copy the topic and you get a worse video about the same thing. Copy the mechanism and you get a fresh video that wins.

So I built an AI system that pulls competitor videos apart, finds the mechanism inside each one, and remixes those mechanisms into ideas that fit your brand. Then it ranks them. Then it writes the scripts.

It's called Short-Form Kit. Six skills. Open source. MIT. Free.

Competitor outliers in. Ranked, on-brand scripts out.

Discover outliers → Decode why each wins → Remix into ideas → Rank → [you pick] → Script

Get it free, open-source, below (scroll to the bottom).

Sponsored

Join 2M+ Professionals Getting Ahead on AI

Keeping up with AI shouldn't feel like a second job.

But between the new tools, viral posts, and endless hot takes, most people spend hours every week trying to figure out what actually matters.

The Rundown AI fixes that. 

It's a free newsletter that gives you the AI news, tools, and tutorials you actually need to know. All in just 5 minutes a day.

Over 2M professionals at companies like Apple, Google, and NASA already read it every morning to stay ahead.

Plus, if you complete the quiz after signing up, they'll recommend the best tools, guides, and courses for your specific job and needs.

Sign up to start learning.

the copy trap

Copying competitor videos feels like research. It isn't. It's the most expensive mistake in short-form.

Here's what actually happens when you copy a winner.

You see a video with 2 million views. You assume the 2 million came from the topic. So you make the same video about the same topic. But you're the second person to make it, the algorithm already spent its appetite for that topic on the original, and you brought none of the things that actually carried it: the hook shape, the pacing, the specific tension it resolved.

You shipped a cover band. Cover bands don't break out.

There's a worse version of this trap, and it's the one that quietly drains budgets. You copy a video that was never an outlier in the first place. It had 2 million views because the creator has 4 million followers and every video they post clears a million. That video underperformed. You just copied a loss and called it research.

That's the move nobody questions:

  ╔══════════════════════════════════════════════════╗
  ║   THE COPY TRAP                                   ║
  ║   ──────────────                                  ║
  ║                                                   ║
  ║   See a video with big absolute views             ║
  ║        │                                          ║
  ║        ▼                                          ║
  ║   Assume the views came from the topic            ║
  ║        │                                          ║
  ║        ▼                                          ║
  ║   Remake the same topic, one week late            ║
  ║        │                                          ║
  ║        ▼                                          ║
  ║   400 views                                       ║
  ║        │                                          ║
  ║        ▼                                          ║
  ║   "short-form doesn't work for us"                ║
  ╚══════════════════════════════════════════════════╝

This isn't a creativity problem. It’s not even a production problem.

Nobody decoded the video before they copied it.

Nobody asked which part of it was repeatable. Nobody checked whether it beat the creator's own baseline or just rode their follower count. Nobody pulled the mechanism out and separated it from the topic. So the calendar got filled with covers of other people's covers, and the feed does not reward that.

The fix is a process. Decode, don't copy.

Twitter tweet

what I built

Short-Form Kit is a six-skill Claude Code plugin that turns competitor short-form videos into ranked, on-brand video ideas and finished scripts.

Three skills you interact with directly, three that run under the hood:

→ The idea engine. The orchestrator. Discovers outlier videos, runs the decode on every one, remixes the parts into ideas, ranks them, and hands the top picks to the scriptwriter.
→ The video decoder. Takes one video and breaks it into its reusable components: the root, the visual hook, the written hook, the spoken hook, and the three extractables. Runs standalone or inside the engine.
→ The script writer. Takes one approved idea plus your brand profile and writes a production-ready script: three-hook block (visual, written, spoken), beat-by-beat body with retention marks, CTA, and a shotlist.

It composes a few tools underneath. Virlo discovers outliers by niche keyword. ScrapeCreators pulls transcripts and cover frames. A Gemini pass adds visual analysis on the strongest outliers. You don't wire any of that by hand. The engine does.

It runs two ways. Keyword mode discovers outlier videos for a niche across Instagram, TikTok, and YouTube. Handles mode analyzes a specific list of competitor handles. Either way, the output is the same: ranked ideas, then scripts.

Works with Claude Code, Codex, OpenClaw, or any agent framework that reads skills.

decode, don't copy

Twitter tweet

This is the framework. It's the thing you steal from this email even if you never download the kit.

You never copy a video. You decode it first. Decoding means classifying a winning video into parts that are actually reusable, then rebuilding from the parts.

Step one: root it.

Before you analyze a video, classify it. A cooking tutorial and a cooking opinion piece share a topic and win through completely different mechanisms. If you skip this step you'll misread why the video worked.

The kit roots every video on a three-level ladder:

→ Level 1. Fiction, Non-fiction, or Lifestyle.
→ Level 2. Domain plus treatment. Tech + commentary, Food + demo, Money + opinion.
→ Level 3. The treatment sub-type: educational, argumentative, opinion-based, narrative, or demo.

An AI news short about a model launch roots to Non-fiction > Tech + commentary > opinion-based. A "build this in 60 seconds" video roots to Non-fiction > Tech + demo > demo. Same niche, different roots, different win conditions. Root first or you remix blind.

Step two: pull the three parts.

Every decoded video yields exactly three reusable components. Keeping them separate is the whole game.

The Viral Vector is what most people miss. It's not the topic. A curiosity gap in the first two seconds is a Viral Vector. "Counterintuitive claim stated as fact, proven inside the video" is a Viral Vector. A visual before/after that needs no narration is a Viral Vector. Those transplant onto any topic. The topic itself does not.

Copy the topic, ship a cover band. Borrow the Viral Vector, ship something new that runs on a proven engine.

the four rings

Here's where the fresh ideas actually come from.

Most operators only watch their direct competitors. That's exactly why their content looks like everyone else's. Direct competitors give you saturated ideas. Everyone in your niche is already watching the same five accounts and copying the same five videos.

The kit maps competitors into four rings:

  ┌─────────────────────────────────────────────────────┐
  │              THE FOUR COMPETITOR RINGS                │
  ├───────────────┬───────────────────────────────────────┤
  │ DIRECT        │ Same niche, format, audience.         │
  │               │ → saturated. Low signal.              │
  ├───────────────┼───────────────────────────────────────┤
  │ INDIRECT      │ Same audience, different topic.       │
  │               │ → borrow the TOPIC from here.         │
  ├───────────────┼───────────────────────────────────────┤
  │ ADJACENT      │ Same format, different niche.         │
  │               │ → borrow the FORMAT from here.        │
  ├───────────────┼───────────────────────────────────────┤
  │ DISTANT       │ Unrelated niche entirely.             │
  │               │ → only for unusually portable Vectors.│
  └───────────────┴───────────────────────────────────────┘

The extraction rule is the part that breaks the copy trap:

→ Viral Vectors are valid from any ring. A great mechanism is a great mechanism, no matter where you found it.
→ Interest Topics carry the most signal from indirect competitors. They share your audience, so the topics winning for them tell you what your audience is hungry for right now.
→ Formats carry the most signal from adjacent competitors. A format winning in a different niche proves the container works, divorced from any one topic. Bring it into your niche and you're early.
→ Direct competitors mostly give you format parity you already have. Low signal. Watch them for Vectors only.

Familiar but new. You borrow a format from outside your niche, a topic from inside it, and a mechanism from anywhere. The result feels native to your audience and looks nothing like what your direct competitors are shipping.

That's the difference between research and copying.

an outlier is not a big number

One more piece, because the kit refuses to get this wrong and most operators get it wrong every time.

An outlier is a video that beats its own channel's baseline. Not a video with big absolute views.

A channel that normally does 50K views posts one that does 500K. That 500K video is a 10x outlier. Real signal. Something specific carried it.

A channel that normally does 4M views posts one that does 2M. Two million views looks huge. It's a loss. The creator's own audience showed up at half the usual rate. There's nothing to learn from it except what not to do.

The kit measures every video against the creator's own recent median. In handles mode it pulls roughly 20 recent posts per creator and keeps only the ones beating that creator's median by 2x or more. In keyword mode it discovers videos already flagged as outliers relative to their channel. Every decode carries an outlier magnitude, 8x baseline or 3x baseline, and that ratio feeds the ranking.

Absolute views lie. Outlier magnitude tells the truth. Decode the 10x video, ignore the 2M-view loss.

from outliers to scripts

The engine runs a ten-step pipeline. You talk to it in plain English. It handles the steps.

  ┌─────────────────────────────────────────────────────┐
  │              SHORT-FORM KIT PIPELINE                  │
  │                                                       │
  │   0.  Intake (mode, brand profile, run slug)          │
  │   1.  Discovery (find outliers, keyword or handles)   │
  │   2.  Receipts (pull transcripts + cover frames)      │
  │   3.  Decode (root + 3 hooks + 3 extractables)        │
  │   4.  Deep pass (Gemini visual analysis, top 5-8)     │
  │   5.  Ideate (10-15 remixed, brand-fit ideas)         │
  │   6.  Rank (viral potential × brand fit)              │
  │   --- GATE: you pick which ideas to script ---        │
  │   7.  Script (one production-ready script per pick)   │
  │   8.  Log (every claim traced to a raw file)          │
  │                                                       │
  └─────────────────────────────────────────────────────┘

A few things worth slowing down on.

The decode is structured. Every video becomes a Decode Record with 16 fields: root, visual hook, written hook, spoken hook, the Viral Vectors, the Interest Topic and its type, the Format, the competitor ring, the outlier magnitude. Nothing is a guess. If the kit can't decode a video, it skips it and logs why instead of faking a record.

The deep pass is honest about what it knows. The top 5 to 8 outliers by magnitude get a Gemini visual analysis: pacing, cut rhythm, visual style, retention devices. Those decodes get tagged deep. The rest stay shallow, and any idea built on a shallow decode is flagged in the ranked table. The kit never pretends it analyzed footage it didn't.

Ideation borrows, it doesn't invent. Stage 6 pools every Viral Vector, Interest Topic, and Format from every decode and remixes them into 10 to 15 ideas. Each idea names exactly which Vector, Topic, and Format it borrowed, and states the brand angle. If the decode pool is too thin to support 10 real ideas, the kit makes fewer and says so. It will not fabricate a mechanism that wasn't decoded.

ranked before you script

This is the part that separates a kit from a content firehose.

Most AI tools hand you 30 ideas. Thirty ideas isn't a plan. It's a second blank calendar.

Short-Form Kit scores every idea on two axes before it writes a single script.

Viral potential (1 to 10). The average of five sub-factors: Viral Vector strength, outlier magnitude, format repeatability, topic freshness, and hook strength.

Brand fit (1 to 10). The average of four sub-factors: voice match, audience match, content-pillar match, and authentic credibility.

Then it combines them:

combined = viral_potential × (brand_fit / 10)

Brand fit is a multiplier, not a tiebreaker. A viral-10 idea that scores 4 on brand fit lands at 4.0. A viral-7 idea that scores 9 on brand fit lands at 6.3. The on-brand idea wins. That math is deliberate. A video that goes viral for the wrong audience isn't a win, it's a distraction, and the rubric refuses to rank it like a win.

There's also a hard gate. Before any scoring, every idea is checked against your brand's anti-topics, the subjects you have decided you will never make content about. A match is an instant disqualification. A 10-viral idea on an anti-topic is still dropped, listed separately, and never scored. The kit will not tempt you with off-limits content because the numbers looked good.

Then the real gate. The kit presents the ranked table and stops. It does not script anything until you pick. Default suggestion is the top 3 to 5. You choose. You stay the editor. The agent is the research and drafting engine, not the decision-maker.

Once you pick, the script writer takes each idea and writes it: a three-hook block (visual hook, written hook, and spoken hook) where each complements instead of repeats the others, a beat-by-beat body, retention marks at every drop-off point (open loop, pattern break, callback, escalation, visual reset), a CTA in your voice, and a shotlist. Ready to shoot.

what this agent refuses to do

The rules baked into every stage.

→ Never call a big number an outlier. Outlier magnitude is measured against the creator's own baseline. A high-view loss is a loss.
→ Never invent an extractable. Ideas only remix Viral Vectors, Topics, and Formats that were actually decoded from real videos. Thin pool, fewer ideas, stated plainly.
→ Never infer footage it didn't watch. Shallow decodes are labeled shallow. Visual claims only appear when the Gemini pass actually ran.
→ Never script past the gate. You pick the ideas. The kit waits.
→ Never lose the receipts. Every view count and outlier ratio in any artifact traces back to a raw API response saved on disk. The run folder is auditable end to end.

If you have ever had a content strategy doc fall apart because someone caught one made-up number, these are the rules that stop that.

bad ideas cost the same as good ones

Twitter tweet

Here's the math nobody runs.

A video built on a copied idea and a video built on a decoded idea cost you exactly the same to make. Same shoot. Same edit. Same hour out of your week. The production bill does not care whether the idea behind the video was any good.

So the idea is the only variable that actually moves your numbers.

SAME COST TO SHIP                    DIFFERENT OUTCOME
─────────────────────────────────    ─────────────────────────────────
Copied idea                          Decoded idea
  Filming + edit:     ~5 hrs         Filming + edit:     ~5 hrs
  Cost to produce:    identical      Cost to produce: identical
────────────────────────────       ────────────────────────────
Runs on a stale topic                Runs on a proven mechanism
→ 400 views, then gone               → real shot at an outlier

You're already paying to make the videos. The expensive part was never the kit or the research. The expensive part is shooting a week of content on ideas that were never going to land.

Short-Form Kit is free. The research it runs costs pennies per video. The only thing it changes is whether the idea behind each video was decoded or guessed. Across a month of posting, across a roster of brands, that one change is the whole growth curve.

who this is for

Run this kit if:

→ You ship short-form video and the calendar is always hungry
→ You're an agency or operator running video for more than one brand
→ You keep copying competitor videos and keep wondering why the copies die
→ You have a brand you can articulate (voice, audience, content pillars, anti-topics) and want ideas that actually fit it
→ You're already running Slideshow Kit or Creator Breakout Kit and want the video-ideation engine that feeds them

Not for you if:

→ You're a solo creator with one channel and a deep instinct for your own niche. Keep trusting it, you're winning
→ You already have a senior short-form strategist running post-level outlier analysis and rejecting off-brand ideas in writing. You don't need this
→ You want a tool that hands you 30 ideas and calls it a strategy. That's the second blank calendar, and it's not what this is

get the kit

Full system free on GitHub:

You get:

→ The idea engine that runs the full ten-step pipeline from competitor discovery to finished scripts
→ The video decoder that breaks any short-form video into a 16-field Decode Record (root, visual hook, written hook, spoken hook, the three extractables, outlier magnitude)
→ The script writer that turns an approved idea into a production-ready script (three-hook block with visual, written, and spoken hooks; beat-by-beat body; retention marks; CTA; shotlist)
→ The decode-don't-copy framework: rooting ladder, Viral Vector / Interest Topic / Format split, and the four competitor rings
→ Two discovery modes: keyword mode across Instagram, TikTok, and YouTube, and handles mode for a specific competitor list
→ Outlier logic that measures every video against its own channel baseline, not absolute views
→ A ranking rubric that scores viral potential against brand fit, with an anti-topic hard gate
→ An approval gate so you pick which ideas get scripted before any script is written
→ An honest deep pass that labels shallow vs deep decodes and never claims visual analysis it didn't run
→ A traceable run folder where every view count and outlier ratio traces back to a saved raw API response
→ Graceful degradation so a missing key disables one mode instead of breaking the whole engine

Runs on Claude Code, OpenClaw, or any agent framework that reads skills. MIT license. Fork it.

No upsell. No catch.

Stop copying competitor videos. Start decoding them.

Go big,

Matt

P.S. Short-Form Kit is the video-ideation engine. If you're already running Slideshow Kit, Creator Breakout Kit, Brand Shoot Kit, or Outcome Kit, this is the kit that decides what short-form video to make in the first place. Creator Breakout sets the creator strategy. Short-Form Kit decodes the outliers and writes the scripts. Slideshow Kit ships the carousels. Brand Shoot makes the visuals. Outcome Kit tells you which ones worked.

P.P.S. Star the Short-Form Kit repo if this helps. It tells me to keep building.

The $12k seat I cut from my agency this quarter wasn't a designer. It wasn't an editor. It was a content lead… and the thing that replaced them costs me about $30 a month.

It's called Slideshow Kit. Open source. MIT. Free. Link at the bottom.

It runs daily social on autopilot. Reads your brand's DNA, watches what's trending today, ranks concepts before it writes a single slide, picks a format with real rules, drafts the script, directs the visuals slide-by-slide, and ships through Postiz. Drafts by default. Autopilot once you've earned it.

Today's trends → Brand DNA lens → Ranked concepts → Format-locked script → Scene direction → Render → Postiz → Ship

Trend signal in. Point of view applied. Save-worthy carousels out, every day.

Marketing isn't a department anymore. It's engineering. And the calendar finally broke under the weight of that fact.

Sponsored

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

Why I had to build it

Your feeds are hungry every morning.

LinkedIn wants daily. Instagram wants daily. TikTok wants daily. X wants hourly. Threads. Bluesky. The list keeps growing and the calendar doesn't. Distribution is the moat, and the moat needs filling every single day.

So operators turn to AI. And here's where it goes sideways.

Most AI carousels in the feed right now look like every other AI carousel. Six slides. Stock backgrounds. The kind of post you scroll past without remembering it was there. Technically correct. Emotionally dead. The model has nothing to pull from except the average of everything.

I almost killed this project in March for exactly that reason. The first runs spat out the same generic boring stuff every other AI tool ships. I had to fix it before it was worth releasing. And the fix wasn't a better prompt or a smarter model. The fix was teaching the agent my taste.

Let me back up.

The carousel I tried to copy

A few years ago I was scrolling the feed and Oren John posted a carousel.

Glorious design. Beautiful layout. The kind of post you stop on. I tried to make one myself. Took me hours. Mine looked like trash next to his. So I started a swipe file. Dickie Bush. Other creators with signatures you could spot from a thumbnail before reading a word. I kept saving them.

Instagram post

I obsessed over what made those posts different. It wasn't the writing. It wasn't even the design system. It was taste, a thousand small decisions layered on top of each other until the post felt inevitable. You can't prompt taste. You have to build it.

Last year I built an internal tool to automate carousels for our agency. Outputs came out great. Millions of views. The brands running on it felt the lift. Creative testing and volume matter more than ever, and we were finally keeping up.

Then a few months ago I saw other operators jumping into the carousel game with their own agentic systems. I liked what I saw. I didn't love it. I thought I could make something cooler.

But the v1 had a flaw I couldn't fix without help. It relied on the user to know what was happening in their niche, what was trending, what people were actually talking about right now. I can't ask one person to know everything popping off across millions of conversations. The world moves faster than any one person can read. Models train on stale data. The operator is one human doing too much.

So I built the gap-closer. That's the kit you saw above.

the lane humans can still win

A real personality video can't be faked (and you shouldn’t). If your founder can sit on camera and sound like a founder, that's a moat. Don't AI-UGC over that.

But you can't film one human 15 times a week, testing hooks, formats, and everything else. The video lane has a hard human ceiling and the feeds don't care.

That leaves the volume problem nobody owns. Thousands of posts. Hundreds of accounts. Dozens of channels. We only have so many hours in a day.

Carousels win this lane. You can ship them. You can test them. You can still tell stories. The reader takes their time, scrolls at their own pace, and listens to a podcast in the background while they read. Your carousel doesn't fight for their AirPods. That's a different deal than a video makes with a feed.

The volume war is rigged against humans. Carousels are how humans win it back.

Get it free, open-source, below (scroll to the bottom).

two stacked layers, not one blob

The kit thinks about a carousel as two stacked layers:

You mix the two however you want. Same brand, same style, seven different format moves depending on what today's trend deserves.

The seven formats:

The operator picks which format runs per trend per day, or just tells the agent what they want and lets it decide.

YOU are the storyteller. the agent's the engine.

Most AI carousel tools want to be the storyteller. They generate the angle, the hook, the slides, the captions. The output is technically right but emotionally dead because, say it with me, AI wants to default to mediocre.

Most carousels you scroll past spit out facts over generic stock backgrounds. And to their credit, some get views. Do they convert? I don't know. Do they make somebody fall in love with the brand? Almost never.

I'm not here to recite Wiki facts. I'm here to make somebody feel something.

The agent can't always figure out the story for you. But if you have a story to tell, it can be the engine that tells that story on loop. That's a different product. A taste-curated engine, not a storyteller. You stay in charge of what gets said.

The whole work is teaching the agent your taste.

When I built the early versions in OpenClaw, the first run made the same generic stuff every other AI tool makes. So I asked it for ten versions. I gave feedback on every single one. This one. Not this one. This is too on-the-nose. This is closer. Make ten more in the direction of slide 4. Then again. Then again.

That feedback loop is what builds the kit. Not a prompt. Not a model. Not a pipeline. The taste.

Slideshow Kit ships with a brand DNA contract that turns that taste loop into something portable. Four files per brand:

The kit reads those files directly. Every sub-skill does too. The brand state never gets duplicated or vendored inside the kit. Your voice stays yours. The kit is the engine that runs your voice on loop, not a vendor that holds it hostage.

Concepts before slides

Here's the part most AI carousel tools skip, and it's the part that decides whether anyone saves the post.

Most tools write slides from a prompt. Slideshow Kit ranks concepts first, then writes slides from the winning concept.

Each concept comes with a format pick, an arc, a close action, multiple hook variants, a visual hook, and a save-worthiness score. You pick the concept. Then it drafts. You don't get one shot at a slide deck. You get a slate.

Two details that matter more than they sound:

→ Each concept carries a personal-fact risk flag. Autopilot skips concepts that would put a specific claim about you in your mouth. The model never decides on its own that you "raised $4M" or "left McKinsey." If it can't verify it from your DNA files, it won't ship it.

→ The format and the concept are locked before the slides get written. So the post can't quietly drift from "5 tells" to "6 tells" to "and here's a story" by slide 4. The format has slot rules. The lint chain enforces them.

visuals are planned, not decorated

The kit doesn't go straight from script to image.

Between the script and the render sits a scene director. One small LLM call that writes a specific visual brief per slide, based on the script, your style spec, and the character profile if you're running a character-driven look. The render step gets a directive, not a guess.

That's why six slides in a kit run actually feel like one post instead of six unrelated images that happen to share a palette. The visuals carry the argument. They aren't ornaments stapled onto words.

The first three runs we did across one of our clients tanked their bookmark rate. The scene director didn't exist yet. The lint chain didn't exist yet. Slides were technically correct and visually disconnected. That's why both exist now.

Failure is the spec sheet. Every guardrail in this kit is a scar.

your signature look, on every post

Brand DNA tells the kit how you talk and what you believe. Format tells it what shape the post takes. There's another layer on top of that, what your content looks like in the feed.

Styles. Your own signature look. The visual personality that makes your carousels recognizable from a thumbnail before anyone reads a word. Dickie Bush has one. You can spot a Dickie Bush carousel from across the feed without seeing the face. That's a recognizable style.

The kit lets you build yours and run it on every post that comes after. Drop a few carousels you love into a folder. Tell the agent "make mine feel like that." It writes the spec, saves the references, and from then on every carousel you ship runs through that style. Teach it once. Yours forever.

How it works:

Twitter tweet

The "branded carousel" and "social-native, iPhone-real, recurring-character" looks I used to ship as fixed formats are now just two example styles. You can build your own. You can build five.

save-worthy or it doesn't ship

Bookmark rate is the leading metric on every thread, carousel, and article we ship. Views are vanity. Saves are intent.

So the kit runs a save-worthiness filter on every draft before it leaves the loop. If a draft doesn't read like something a real human in your ICP would save, the kit flags it before Postiz ever sees it.

the daily loop

If you've made it this far, you want to see the orchestration. Most AI tools show you the output and hide the wiring. Here's mine.

┌─────────────────────────────────────────────────────┐
  │              SLIDESHOW KIT DAILY LOOP                 │
  │                                                       │
  │   1.  Load brand DNA                                  │
  │   2.  Live-spend gate (drafts default)                │
  │   3.  Pull today's trends                             │
  │   4.  Daily check-in (Telegram or in-session)         │
  │   5.  Generate ranked concepts + hook variants        │
  │   6.  Pick concept, draft format-locked script        │
  │   7.  Write per-slide scene direction                 │
  │   8.  Render configured styles (PNG)                  │
  │   9.  Build Postiz payloads                           │
  │  10.  Log the run + increment run count               │
  │                                                       │
  └─────────────────────────────────────────────────────┘

Run it manually by asking your agent. Run it on a schedule (a cron) at 9am every day. Run it across a dozen brands in a single shell loop. The workflow is markdown. The host agent reads it (Claude Code, Codex, OpenClaw, Hermes). You can also run it as a plain local checkout with no host agent.

Twitter tweet

Two pieces of plumbing matter more than they sound. The Telegram bot wrapper and the Postiz wrapper are decoupled with distinct exit codes. A Telegram outage never blocks publishing. A Postiz outage never blocks the alert. Per-slide rendering is resilient too. If one slide fails, the loop logs it, keeps going, and writes the partial output. You see exactly which slide failed and why instead of losing the whole pack to a single timeout.

The loop ships even when one rail is down.

Postiz earned its position

Slideshow Kit's publishing layer composes Postiz. Open-source social scheduling that fans out across LinkedIn, X, Instagram, Threads, TikTok, YouTube, Bluesky, Mastodon, the works. One CLI. Every channel that matters.

Worth knowing the backstory because it's the move every operator reading this should study.

Nevo built Postiz, shipped through years of slow growth, stuck around $2-4k MRR. Kept shipping. Came back from side projects. Kept shipping. Then the n8n and social-automation wave hit and MRR finally moved. Then agents hit. Within days he had a CLI and an agent surface ready to go. Big X articles dropped. He took off.

He's clearing about $90k MRR now. ~$1.08M ARR. One-man operation. Open source. I run a copy on my Synology at home.

Twitter tweet

Ship every week for years through slow growth, and when the wave finally arrives, you're ready for it. Watching builders build is one of the best things in this game.

The kit drops carousels into Postiz as drafts by default. One approve click fans them out across every channel you've connected. Autopilot ships them scheduled when you've earned the run history.

If you build marketing systems and you're not running postiz, you're paying SaaS rent for something open source does better. Run it. Star the repo. Send Nevo something nice on X.

the cost flip

If you've made it this far you're probably doing the math in your head. Let me do it for you.

THE OLD WAY                          THE AGENT WAY
─────────────────────────────────  -----------------------
Content lead:           $12,000      Slideshow Kit:   free
Designer (carousels):    $3,000      Image gen: ~$0.10/slide
Time per carousel:      2-4 hours    Postiz:      free
                                     Time /carousel: minutes
─────────────────────────────────  ────────────────────────────
ONE BRAND, ONE MONTH:  ~$15,000      ONE BRAND/ONE MONTH: ~$30

A real content lead still wins for the centerpiece campaigns. Nothing replaces a senior writer when the post is the centerpiece.

But the centerpiece is 10% of your feed. The other 90% is daily LinkedIn, weekly Instagram, the carousel pile your TikTok needs to stay interesting, the brand-pillar reminders, the IG variant. That work is what this kit eats.

For an agency, the math compounds across brands. We've gone from N to 20-50N output in the last twelve months. That's not a marketing claim. That's the line item.

Field notes

Four things I've learned shipping this on real feeds. The kit doesn't teach you these. The reps do.

→ Saves > views. Trust the lint chain. If it flags a draft three times in a row and you ship anyway, your bookmark rate will tell you the same thing the kit already told you.

→ TikTok wants drafts. Don't auto-publish to TikTok from the API. Push as drafts and finish the work in-app.

→ Warm up new accounts before you ship volume. A brand new account cranking out 5x daily looks like a bot to every algorithm. Build human-shaped activity for two weeks first. Jack Frick on X has the right primer.

→ Eat your own dogfood for at least seven runs. That's also the autopilot gate. The kit won't let a brand spend or schedule unattended until it has seven runs of draft history.

who this is for

Run this kit if:

→ You're an agency or operator running social for more than one brand and the calendar math broke

→ You're a founder who knows distribution is the moat and you can't keep up by hand

→ You've got brand DNA you can articulate (voice, perspective, visual system) and you want it run on loop → You're already running Brand Shoot Kit, Creator Breakout Kit, or Outcome Kit and need a daily distribution engine that ships what those kits produce

Not for you if:

→ You're a solo creator with one channel and one human voice you record daily. Keep doing that, you're winning

→ You already have a senior content lead writing five carousels a week with great taste and a tight feedback loop. Keep them

→ You want a carousel toy that ships ten viral posts on day one without you giving it your taste. That's the average AI does on its own, and it's not what this is

get the kit

Full system free on GitHub:

The installer detects your installed agents (Claude Code, Codex, OpenClaw, Hermes) and installs the kit to each.

Three clusters of what you're getting:

THE ENGINE → The daily-loop orchestrator (trends → DNA lens → concepts → script → scene direction → render → postiz) → A concept generator that ranks angles, drafts hook variants, scores save-worthiness, and flags personal-fact risk before any slide gets written → Seven format-locked carousel structures with per-slot lint rules → A scene director that writes one visual brief per slide before render

THE TASTE LAYER → The brand DNA contract (voice, perspective, visual system) plus a corpus loader that auto-finds notes you've already written elsewhere → The styles system — your signature look, authored conversationally, run on every post → A lint chain with a save-worthiness filter so low-bookmark-rate drafts get flagged before postiz sees them

THE PLUMBING → Live-spend gating (drafts default, autopilot opt-in after seven runs per brand) → Decoupled telegram + postiz wrappers (independent failure domains) → Resilient per-slide rendering, multi-brand workspaces, cross-agent install → A structured run packet on disk so every run is auditable → A smoke-eval harness and a built-in health check

You'll need an OPENAI_API_KEY for live image generation, postiz CLI auth for publishing, and optionally a Telegram bot token for the morning check-in. Dry-run path needs no keys.

Setup is about an afternoon. Paste five to ten of your best posts and the kit extracts your DNA. Run a draft-mode loop tomorrow morning. Ship from postiz with one approve click. Watch the output for a week. Then flip to autopilot.

Marketing isn't a department anymore. It's engineering. You're not just building product. You're building the marketing engine that ships it.

It's not about manually pushing pixels. It's about managing the systems that push the pixels for you.

If your feed is hungry every morning, this is the kit that feeds it.

Go big,

Matt

P.S. Slideshow Kit is the daily distribution engine. If you're already running Brand Shoot Kit, Creator Breakout Kit, Outcome Kit, Landing Page Factory, or SEO Kit, this is the kit that takes what they produce and ships it to your feeds every day. Brand Shoot makes the visuals. Creator Breakout decides the angles. Outcome Kit tells you which ones work. Slideshow Kit posts them.

P.P.S. Run postiz. Star the repo. Send Nevo something nice on X.

P.P.P.S. Star the Slideshow Kit repo if this helps. It tells me to keep building.

This single AI tool just made the ten-thousand-dollar studio photo shoot completely obsolete.

If you run an e-commerce brand, you know that photos sell the product. But getting fresh, high-quality shots for your site, socials, and emails costs a fortune and takes weeks. And until now, AI image generators couldn't read your packaging without melting the text into nonsense.

But OpenAI just dropped GPT Image 2, and it changes everything.

I built a free, open-source tool called the Brand Shoot Kit. In this article, I’m going to show you how to drop in a single URL, and 10 minutes later, get a complete 12-shot campaign for a dollar fifty. Let me show you how it works.

A new king is crowned

For the last several months, Nano Banana has been the undisputed king of AI image generation. Every product shot, every brand asset, every "look what AI can do now" tweet - Nano Banana.

But with GPT-Image-2 (by OpenAI) a new winner just got crowned.

Twitter tweet

Twitter tweet

GPT Image 2 holds package geometry. It renders label text without melting it into nonsense glyphs. It respects ratio. It does not hallucinate a third can of supplements where there should be two.

For ecommerce, that is the whole game.

Twitter tweet

PHOTOS ARE THE PRODUCT

I work with a lot of ecomm brands. The thing every single one of them needs more of, all the time, forever, is killer photographs.

Photos sell the product. Photos make it feel real. Photos make it sexy.

The brands that ship more product imagery, more often, in more contexts, on more channels, win. The brands that don't, lose.

That’s the part nobody likes hearing. You can have the best supplement, the best skincare line, the best coffee on earth but if your PDP has one tired hero shot from a 2021 studio shoot, you’re leaving money on the table every minute the page is live.

So when GPT Image 2 dropped, I built an agent kit around it.

It is called Brand Shoot Kit. Open source. MIT licensed. Free.

Paste a product URL. Get back a 12-shot ecommerce shoot. PDP, lifestyle, social, email, marketplace. With QA. With reroll. With channel exports. With a static review frontend you can ship from.

Website in. Killer photos out.

URL → Scout → Plan → Generate → QA → Reroll → Export → Frontend

Let me show you what comes out.

Sponsored

I started using Wispr Flow last week and am loving it so far. Thanks to Wispr Flow for sponsoring this email. Great product.

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

THE CATALOG GAP

Wen you have 50 SKUs, your business carries what I call 'The Catalog Gap.'

A photographer can only do so many shots a day.

So you fake it.

Stock photo on the blog. Last summer's hero on the email. The Amazon listing has 2024 packaging because nobody pulled the new bag through the studio. The seasonal variant ships in two weeks and there are no shots yet. The PDP has one product image when the conversion data says it needs four.

Every ecom brand carries this gap.

Until now, the only fix was hiring more photographers. With GPT Image 2 holding product fidelity at a level Nano Banana never could, the math finally works.

Thats what this kit is for.

WHAT COMES OUT OF THE BOX

You paste a Shopify URL. Ten minutes later, you have:

→ a hero shot at 1024 × 1024 for the PDP
→ four secondaries at 1024 × 1280 (label detail, ingredient pour, hand-held scale, in-context use)
→ two lifestyle frames at 1920 × 1080 (kitchen counter, gym bag, bathroom shelf, whatever the category needs)
→ two social cuts at 1080 × 1920 for Reels, TikTok, Stories
→ an email hero at 1920 × 1080 that doesn't look like every other email hero
→ a marketplace square cleared for Amazon and Shopify Collabs
→ a bundle layout if the product ships as a kit

12 frames. Ratio-aware. Each one rendered at the dimensions the channel actually wants, not stretched in post.

These aren't "AI-looking" product images. They look shot. Real lighting. Real depth of field. Real label fidelity. The kind of frames you'd expect back from a studio day, except you got them in 30 minutes from a URL.

Every shot scored against a 6-criteria QA rubric. Every failure rerolled with the same product reference and a rewritten prompt. Every export packaged up nice and pretty.

THE LIVE PROOF

I ran this on real brands across four categories: Supplements, Skincare, Cleaning Kits, and Coffee.

Perfect passes on all four.

The system deterministicially holds the label text and respecting packaging geometry. The coffee bag artwork renders correctly.

The supplement kits keep the right number of bottles.

New categories add a fixture and a few rules. They don’t require touching the pipeline.

THE OLD WAY

I've been on shoots that made an $11k photographer look like a rounding error.

A few years back I worked on Hennessy's Flaunt Your Taste campaign. Pharrell did the music. We had print, digital, and TV all running together. Models, vintage cars, multiple countries. The main piece, I'm pretty sure, was shot somewhere in France. Months of production. A crew you needed a coordinator just to coordinate.

The shots came back beautiful. The campaign cost more than most ecommerce brands will spend on photography in a decade.

Here's what's wild.

We could produce that entire campaign today. Brand Shoot Kit gets you 90% of the way there on stills. Seedance and the new video models cover the last 10% on motion.

Six months and a global crew, run from a laptop in a day.

That math just changed everything.

THE COST FLIP

Even at the small end of this, the numbers are wild.

A real photo day still wins for the once-a-quarter centerpiece. Nothing replaces a stylist on set when the shoot is the centerpiece.

But the centerpiece is 10% of your visual budget. The other 90% (PDP secondaries, label detail, lifestyle variants, email heroes, marketplace listings, blog imagery, seasonal packs, bundle layouts) is the work this kit eats.

$11K to $1.50 means a brand running 50 SKUs can refresh the entire catalog three times a year for less than the catering bill on one shoot day.

THE MAGIC MOMENT

Every run produces this:

A static, offline-safe gallery at the packet root. Drop the folder anywhere. The page works.

Open it and you see the full shoot. Twelve frames in a premium grid. Hover any one and you get the QA score, the requested ratio, the final dimensions, the channel it was rendered for, and a thumbnail of the reference image it was generated from.

Above the grid, a hero strip with the brand, the run status, and the approve / reroll / reject summary. Below it, the export links by channel (PDP, lifestyle, social, email, marketplace) already rendered at the dimensions each one needs.

That’s what you show a brand manager.

Most AI image tools end at "here is a folder of PNGs, good luck." This one ends at a review surface a human ships from.

THE NON-NEGOTIABLES

Four rules baked into every stage.

1. Product truth beats aesthetics. A pretty image of the wrong package is a failure.

2. Real references only. No story art. No tout art. No logos masquerading as packaging. Real product imagery or the run does not start.

3. QA is the gate, not a suggestion. Failed shots reroll. They do not pass.

4. Live spend is gated. Dry-run by default. Capped shot count. No accidental bills.

WHAT THE SHOOT BECOMES

The PDP hero. The secondary. The label detail. The hand-held scale shot. The lifestyle frame. The email hero. The marketplace square. The bundle layout. The seasonal variant. The blog header. The Reels cover. The Shopify Collabs asset.

Every channel that needs a fresh product image, fed.

WHO THIS IS FOR

You should run this kit if:

→ You sell physical product on a website you control
→ You have 10+ SKUs and a real catalog
→ Your photographer pipeline delivers a fraction of what your website actually needs
→ You ship on Shopify, WooCommerce, or anything with structured product data
→ You sell on marketplaces and your listings are out of date
→ You run an email program and the heroes go stale every six weeks
→ You launch seasonal variants, bundles, or limited drops faster than you can shoot them

Not for you if:

→ You are a service business with no physical product
→ Your shoot pipeline already produces 400+ shots a month and you are happy with it (in which case, congratulations, also you are lying)

GET THE KIT

GET BRAND SHOOT KIT (FREE)

You get:

→ 8 cooperating skills under skills/
→ 10-stage pipeline from URL to review frontend
→ Ratio-aware generation across 1:1, 4:5, 9:16, 16:9
→ 6-criteria QA rubric with weighted scoring
→ Reroll loop with rewritten prompts and the same product reference
→ Channel-ready export packager with manifest metadata
→ Review frontend at every packet root
→ Deterministic fixtures and golden runs for coffee, skincare, supplements, cleaning kits
→ ./doctor.sh, ./evals/run.py, full live-proof playbook

You will need an OPENAI_API_KEY. The kit runs end-to-end on GPT Image 2 by default and ships a dry-run path that requires no keys at all.

No catch. No upsell. MIT license. Fork it.

Repo: github.com/TheMattBerman/brand-shoot-kit 

Setup is around 30 minutes. Run a no-spend dry proof first. Once index.html reads cleanly on a 3-shot run, drop the cap and run the full 12.

WHAT TO DO ONCE YOUR CATALOG IS FULL

A Brand Shoot Kit run gives you a fresh, on-brand product image library that every other channel is hungry for. Once you have it, the obvious next moves:

→ Plug the shoot into StealAds and let it build paid ad creative on top of brand-true product imagery. The thing that kills AI ads is package drift. You skip that problem entirely. Early access is open at stealads.ai (or reply ads and I will route you in).

→ Run SEO Kit on the freshly refreshed PDPs. Better imagery + better content = rankings move.

→ Use Creator Breakout Kit to turn the same product truth into creator briefs and short-form concepts.

→ Run ScrollClaw to convert customer reviews into UGC-style organic content using the same product references.

→ Use Outcome Kit to figure out which images, ads, and pages are actually doing the work.

Brand Shoot Kit feeds StealAds, SEO Kit, Creator Breakout Kit, ScrollClaw, and Outcome Kit.

If your website is hungry for visuals, this is the one to run first.

Go big,

Matt

P.S. We have been running these kits for ecom brands on their actual catalogs. If you want my team to run Brand Shoot Kit on your store, full catalog refresh, on a cadence, reply to this email and say shoot. I am opening a handful of slots a week.

P.P.S. If you missed the last few kits, they are all still free and they plug into each other: LinkedIn Ads Kit, Creator Breakout Kit, Outcome Kit, Landing Page Factory, ScrollClaw, SEO Kit, Google Ads Copilot, Meta Ads Kit, First 1000 Kit. Brand Shoot Kit is the production lane that feeds every channel kit above it.

You are probably burning thousands of dollars on LinkedIn Ads right now, and you don’t even realize it

Your dashboard says your Cost Per Lead is dropping, and your marketing team is out celebrating.

But when your sales team actually calls those leads? It’s just college students doing homework and random vendors checking out the competition.

I call this the 'Fake Lead Tax,' and it is destroying B2B marketing budgets.

But what if you had an AI operator that caught this daily, separating the real buyers from the fake signal, before you wasted another dime?

Today, I'm giving you the exact, open-source AI agent that does exactly that.

Get it free (scroll to the bottom).

Sponsored

SaaS Billing Is Breaking. Here's How to Fix It.

Usage-based pricing is the new default — but most billing platforms weren't built for it. Get the Tabs Buyer's Guide to learn what modern revenue infrastructure actually requires.

Download the Free Guide.

The fake lead problem (and why LinkedIn makes it worse)

Most LinkedIn Ads teams don't have a cost-per-lead problem.

They have a buyer-quality problem.

Here's how it works. You launch a campaign targeting "Heads of Growth at SaaS companies with 50+ employees." Fine. Then LinkedIn's auto-expansion drifts the audience outward because it wants to spend your budget. Forms get optimized for "easy fill." CPL starts looking great. Leads start flowing.

Except half of them are:

• students doing "market research"

• vendors "checking out the competition"

• solopreneurs at 3-person shops who will never qualify

• enterprise folks who grabbed the deck and will never book a call

• people who already have your competitor and just want the comparison PDF

Sales opens the lead. Sales closes the tab. Nobody follows up. The number on the dashboard is meaningless.

The math is brutal. LinkedIn clicks are $10-25 each. A "cheap" $150 lead is 10+ clicks of wasted spend on the wrong person. If 30% of your leads are fake, you're burning a third of your LinkedIn budget on people who will never buy.

Your dashboard celebrates. Your sales team rolls their eyes.

Same pattern I called out in the Outcome Kit, just LinkedIn-specific. And LinkedIn makes it worse because the clicks are 5 to 10x more expensive than Meta.

The buyer-quality layer never existed.

Nobody checked which leads turned into opportunities. Nobody mapped which message brought the CFO vs. the intern. Nobody decided which Thought Leader post was worth sponsoring based on whether buyers reacted, not whether coworkers liked it.

So the account optimized for the wrong thing. The report looked great. The pipeline stayed empty. And eventually someone on the exec team asked "why are we still spending on LinkedIn" and marketing didn't have an answer.

What I built

LinkedIn Ads Kit is an AI LinkedIn Ads manager that gives you a daily brief: what's real, what's fake, what's leaking, what to do next, and what not to touch yet.

A daily LinkedIn Ads operator with memory and a seatbelt.

The one-liner I keep coming back to:

It runs two ways:

Works with Claude Code, OpenClaw, Hermes, or any agent framework that reads skills.

See the Meta Ads kit in action:

Twitter tweet

The Daily Brief

Start here.

The agent invokes the linkedin-ads skill, reads your brand memory, ingests CSV exports or pulls through the API, and hands back the brief.

If you live in the terminal, the CLI equivalent is:

npm run daily:brief -- --campaigns campaigns.csv --leads leads.csv --crm crm.csv

Or the shorter version:

npm run daily-check

Either way, the brief is read-only by default. You don't have to let the kit touch your account. The brief is useful by itself.

Every brief answers five questions:

1. Are we buying real buyers?

2. What is creating fake confidence?

3. Where is the leak between ad, form, and sales?

4. What should we do today?

5. What should we avoid touching yet?

And it lands in six sections:

The Four LinkedIn Ads States

The daily brief runs on a diagnosis model. Every campaign, every ad, every lead gets sorted into one of four states.

1. Real Signal

The campaign is producing evidence the right buyers care. Qualified leads. Opportunity-stage CRM matches. Comments from actual ICP buyers. Form answers that show real pain.

Operator move: protect the learning. Scale slowly. Create sibling angles. Document why it works.

2. Fake Signal

Campaign looks great in Campaign Manager. Produces weak downstream quality. Cheap CPL. Strong CTR. Students, vendors, tiny companies, job seekers. Sales rejection notes pile up.

Operator move: stop celebrating the dashboard. Tighten audience or form friction. Rewrite the offer around a specific buyer problem. Pause if quality doesn't recover.

3. Trust Gap

Audience and offer are probably right, but the ad has no credibility. Low CTR on company-page ads. Generic product claims. No proof. No human voice. Organic posts outperforming paid creative.

Operator move: test Thought Leader Ads. Add proof. Turn internal POV into external creative. Use a human voice when the company voice feels flat.

4. Handoff Leak

Ad created interest. Lead died after conversion. Sales accepted rate below expectation. Delayed follow-up. Reps lacking ad/form context. CRM stages that don't map back to the campaign.

Operator move: add hidden campaign context. Write sales handoff notes. Add one fit question. Map form answers to sales talking points.

That's the diagnosis. The daily brief is the output: Real Signal gets protected, Fake Signal gets challenged, Trust Gap and Handoff Leak both surface under "Leaks," and every state rolls into "Today's Moves" with a specific operator action.

LinkedIn is not Meta with expensive clicks

LinkedIn is not a volume channel. It's a trust and buyer-quality channel.

Meta rewards creative velocity, emotional hooks, and scroll-stopping patterns. LinkedIn rewards the opposite: the right buyer, hearing a credible point of view, taking a step that filters for fit.

That's why:

• Company-page ads underperform. There's no face. No trust transfer. Logos don't carry conviction.

• Thought Leader Ads work. Trust transfers from the person, not the company.

• Broad audiences drift into garbage. Your buying committee is narrow. LinkedIn optimizes for spend, not fit.

• Cheap leads are a trap. The form that's easy to fill is the form students fill.

LinkedIn Ads Kit is built around this belief. Every stage asks the same question: is this creating a real buyer conversation? Not "did we get a click." Not "is the CPL down." Is there a buyer on the other side who will actually convert?

The 10 skills

Every skill is a plain markdown file. Any agent framework that reads skills can run this. Each skill runs standalone or as part of the full pipeline.

Thought Leader Ads: the sharp edge

Thought Leader Ads are the one format on LinkedIn that consistently beats company-page creative. But most teams sponsor posts because the post got likes.

That's not enough.

This kit scores candidate posts on a buyer-quality rubric:

If API creation is available, the kit drafts the creative action. If LinkedIn's approval mechanics block API creation, it produces a Campaign Manager launch packet: post URL, objective, approval copy, budget test, UTMs, measurement notes, what to watch before scaling.

Either way, you're not sponsoring the post because it got 12 likes from your team's company accounts. You're sponsoring it because the market already told you buyers care.

Safe Apply: the seatbelt

LinkedIn is expensive. A bad change at 2am costs real money. So the kit does not freehand the account.

The loop is:

  READ → BRIEF → DECIDE → DRAFT → APPROVE → APPLY → LEARN

You stay in the driver's seat at DECIDE and APPROVE. The agent takes the grunt work on either side.

So what this looks like for you is something like:

Every apply starts as a markdown draft the agent writes for you. You read it. You approve it. Every draft validates against current account state before a write lands.

V1 supports: pause campaign, pause creative, set campaign daily budget, activate approved draft creative, create Thought Leader creative draft (or Campaign Manager fallback packet).

Still excluded: deleting entities, broad campaign creation, bid strategy changes, audience expansion, blind rebuilds, anything without brand/account validation.

You stay in the loop. The agent takes the drudgery. The account stays un-destroyed at 2am.

Multi-brand memory

If you run more than one brand, this is the part that compounds.

Each brand gets its own memory folder:

workspace/brands/acme/
├── profile.md
├── audience.md
├── voice-profile.md
├── offer.md
├── learnings.md
└── linkedin/
    ├── account.md
    ├── briefs/
    ├── drafts/
    ├── audit-trail/
    └── cache/

The important file is learnings.md. Every brief and every apply run appends what the kit learns about the brand: who converts, what sales rejects, which offers create quality, which Thought Leader voices deserve spend, what not to repeat.

The agent gets smarter every run. The agency gets a durable knowledge base per client. The next LinkedIn Ads manager who rolls onto the account doesn't start from zero.

How this plugs into the rest of the stack

This is the eleventh kit I've open-sourced in 11 weeks (yeah, still counting, and still free). And like the rest, it's designed to fit into the full paid pipeline:

Sign up for StealAds early access here: https://stealads.ai

LinkedIn Ads Kit slots in as the third paid channel operator. If you're already running the Meta Ads Kit and the Google Ads Copilot, this is the missing one.

And if you're already running the Outcome Kit, LinkedIn gets its own channel-specific read instead of being lumped into a generic attribution view. Because LinkedIn is expensive and different. Buyer-quality logic that makes sense for Meta would make you look foolish on LinkedIn.

Who this is for

• B2B founders running LinkedIn paid and tired of explaining cheap-CPL-big-waste to the board

• Agencies managing multiple LinkedIn accounts who need a durable operator framework

• Heads of growth at $1-20M ARR who know LinkedIn should work but haven't cracked the buyer-quality layer

• Operators running Thought Leader Ads and looking for a repeatable way to decide what to sponsor

Not for you if you have a senior LinkedIn operator on payroll who is already mapping leads to CRM stages, scoring posts on ICP fit, and drafting account changes with audit trails. You don't need this. You're the 1%.

Get the kit

Full system free on GitHub:

You get:

• The Daily Brief — read-only by default. Real Signal, Fake Signal, Leaks, Today's Moves, Do Not Touch Yet, Data Gaps. Ask your agent "write me today's LinkedIn brief" (or run npm run daily:brief / npm run daily-check from the CLI).

• 10 skills covering the full pipeline from account read to safe apply

• The Four LinkedIn Ads States framework: real signal / fake signal / trust gap / handoff leak, with a specific operator move for each

• Two modes: Export Mode for zero-setup daily briefs, Connected Mode for API-backed drafts and apply

• Flexible lead form CSV mapping — LinkedIn lets teams rename fields per form, so the kit maps common aliases and preserves unmapped columns. No reformatting required.

• Safe Apply with dry runs, explicit confirmation, current-value validation, and full audit trail. READ → BRIEF → DECIDE → DRAFT → APPROVE → APPLY → LEARN.

• Thought Leader Ads selector with an ICP / pain / trust / organic-signal / offer / sales-usefulness rubric, plus Campaign Manager fallback packets when API creation is gated

• Multi-brand memory that compounds learnings across client accounts (the learnings.md file per brand)

• Form friction review that looks at qualification quality, not just fill rate

• Sales handoff generator that gives reps context for the conversation the ad actually started

• Full CLAUDE.md, skill docs, install script, demo data, the works. Runs on Claude Code, OpenClaw, Hermes, or any agent framework that reads skills. MIT license.

No upsell. Fork it.

Stop paying LinkedIn a fake-lead tax. Start getting the daily brief Campaign Manager should have written for you.

Go big,
Matt

P.S. If you missed the last few kits, they're all still free and they plug into each other: Creator Breakout Kit, Outcome Kit, Landing Page Factory, ScrollClaw, Google Ads Copilot, Meta Ads Kit. LinkedIn Ads Kit slots in next to Meta Ads Kit and Google Ads Copilot as your third paid channel operator.

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

Your creator brief is lying to you.

You're about to drop $40K on sourcing, briefing, and production. Your team spent maybe an hour on the actual idea. That's not a strategy. That's a lottery ticket in a Google Doc.

Look at the doc in your inbox right now. "Authentic UGC. Three hook variants. Five lifestyle creators we like." Nobody pulled the real post-level outliers in your category. Nobody mapped your top buyer objection to a specific hook structure. Nobody rejected the generic plays before briefing.

It doesn't matter if you pay a creator with 10M followers. If the idea is flawed and the hook is generic, people swipe in three seconds. You just paid for production on a concept nobody validated.

So I built a kit that runs the strategy layer before sourcing. Pulls the real outliers in your niche. Maps objections to specific hook structures. Rejects the generic plays in writing, so the test plan has teeth before you spend a dollar.

Open source. Free. Here's how it works.

(scroll to the bottom for the link)

Sponsored

Here’s your lifeline.

Another headline. Another client pays late. The next 10 days shift. You open your bank app before walking into the office.

The hits just keep coming right now.

And as the leader, you’re the one absorbing all of them.

But survival doesn’t come from holding tighter alone.

The Small Business Survivor Guide gives you 83 practical ways to cut costs, stabilize cash flow, and navigate economic pressure with confidence.

Because in times like these, stability isn’t luck. It’s strategy.

And the leaders who stay standing are the ones who prepare for what’s next.

Download The Small Business Survival Guide

The "send the ideas" problem

Every brand and every agency runs the same broken loop.

Step one, you decide you want creator content (or you hire an agency to run it for you). Step two, the "concepts deck" lands in your inbox. Step three, the deck is 80% creator comps someone already liked and 20% generic UGC advice pulled from a swipe file. Step four, the contract gets signed. Step five, sourcing starts. Step six, the brief says "authentic, hooks first, three variants."

If you're already running paid UGC with a system like this AI UGC army or AI posts like a 100k creator, you know the feeling. Execution is no longer the bottleneck. The bottleneck is the idea.

  ╔══════════════════════════════════════════════════╗
  ║   THE "SEND THE IDEAS" LOOP                      ║
  ║   ──────────────────────                         ║
  ║                                                  ║
  ║   Prospect: "send the ideas"                     ║
  ║        │                                         ║
  ║        ▼                                         ║
  ║   Generic UGC deck (creators the team likes)     ║
  ║        │                                         ║
  ║        ▼                                         ║
  ║   $40K on sourcing + briefing + production       ║
  ║        │                                         ║
  ║        ▼                                         ║
  ║   Content built on vibes                         ║
  ║        │                                         ║
  ║        ▼                                         ║
  ║   Underperformance                               ║
  ║        │                                         ║
  ║        ▼                                         ║
  ║   Post-mortem nobody acts on                     ║
  ╚══════════════════════════════════════════════════╝

This isn't a creator problem. This isn't even a briefing problem.

The strategy layer never existed.

Nobody pulled the post-level outliers in the category. Nobody mapped the top buyer objection to a specific hook structure. Nobody decided which generic plays to reject. Nobody ran any of that before sourcing got spun up.

So the sourcing team optimized for the wrong thing. The briefing was generic because the concepts were generic. The creators filmed what looked like every other UGC spot on the platform. And the platform does not reward what looks like every other UGC spot on the platform.

Same philosophy behind the First 1000 Customers kit: decide the strategy layer first, let automation run the execution layer.

What I built

Creator Breakout Kit is an AI strategist that finds the creator concepts, hooks, and angles most likely to break out in a category (before you pay to source or produce a single video).

It runs a 6-stage pipeline. Each stage reads the last stage's artifacts. No stage hallucinates inputs. No stage ships without an honest evidence label.

  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐    ┌──────────────┐
  │ Brand Truth  │───▶│   Breakout   │───▶│   Creator    │───▶│   Concept    │
  │              │    │   Patterns   │    │  Archetypes  │    │    Slate     │
  │  Firecrawl   │    │    Virlo     │    │              │    │              │
  └──────────────┘    └──────────────┘    └──────────────┘    └──────────────┘
                                                                      │
    ┌──────────────┐    ┌──────────────┐    ┌──────────────┐          │
    │   Managed    │◀───│    Brief     │◀───│  First 3     │◀─────────┘
    │ Next Step    │    │  MD + HTML   │    │   Tests      │
    └──────────────┘    └──────────────┘    └──────────────┘

The output is a Creator Breakout Brief. An 8-section document that reads like something a senior creator strategist with a week of research would hand you, except it runs in about 10 minutes.

It ships in two formats from one data source: a markdown file you can edit and feed into other agents, and a dark-terminal-aesthetic HTML file you can open in a browser and actually show a prospect without being embarrassed. Same score logic, same content, one parametric renderer. The formats can't drift from each other.

Works with Claude Code, OpenClaw, or any agent framework that can read skills.

The 8-section Creator Breakout Brief

Every run produces a brief with this exact structure.

Plus an appendix with brand context, hero products, audience, objections, and brand-safety guardrails.

Section 7 is the one that changes how the work feels. Most creator strategy docs are allergic to saying "no." This one explicitly rejects the 5 generic plays before it recommends anything. That single section is why clients read the brief and say "okay, you actually thought about this."

The framework on every concept card

Every single concept in the slate carries four labels. This is what makes the brief feel ownable instead of vibes.

  ┌─────────────────────────────────────────────────────┐
  │              CONCEPT CARD LABELS                     │
  ├─────────────────────┬───────────────────────────────┤
  │  Breakout Driver    │  Which pattern this leans on │
  │                     │  (the thing actually working │
  │                     │  in the category)            │
  ├─────────────────────┼───────────────────────────────┤
  │  Buyer Tension      │  The objection or desire     │
  │                     │  this concept resolves       │
  ├─────────────────────┼───────────────────────────────┤
  │  Creator Delivery   │  Filming style + creator     │
  │  Style              │  energy the concept needs    │
  ├─────────────────────┼───────────────────────────────┤
  │  Test Priority      │  Run first / run alongside  │
  │                     │  / backlog                   │
  └─────────────────────┴───────────────────────────────┘

Then a score strip: Strategic fit · Breakout evidence · Production ease · Confidence. The confidence score is computed from the research mode and whether a real source post is attached. No research → Hypothesis only. Firecrawl + Virlo → Directional. Video intelligence attached → Strong directional.

The agent will not fake certainty. If the evidence isn't there, the label says so. That's the whole game.

The generic creator brief vs. what this produces

Here's the gap, side by side.

Generic creator brief that ships today:

Creator Breakout Kit direction:

The first one buys you production. The second one buys you a test plan.

You could hand the second one to three different sourcing teams and get roughly the same creators back. You could hand the first one to three different sourcing teams and get three completely different decks. That's the signal that the strategy layer is missing.

Why most creator workflows get this backwards

The industry standard is find creators first. Pick based on audience fit or follower count. Brief a generic hook structure. Hope.

This kit flips it.

The creators don't change that much. What changes is what you ask them to do. And what you ask them to do is the entire difference between "this campaign worked" and "this campaign was vibes."

The 7 skills

Every skill is a plain markdown file. Any agent framework that reads skills can run this.

Outlier logic, not follower vanity

This is the part that matters most. And it's the part every other "creator AI" tool gets wrong.

Most tools pull "top creators in the category." That's a follower-count lookup. It tells you who has the biggest audience, not who is making the content that's actually breaking out right now. (This is the same trap I flagged in the viral video kit: the platform doesn't reward follower count. It rewards pattern fit.)

Creator Breakout Kit uses post-level outlier logic. It goes into Virlo, pulls the posts that are outperforming a given creator's own baseline by a factor of 3x, 5x, 10x, and then looks at why. What's the hook shape. What's the mechanic. What's the buyer tension it's resolving. What's the filming style doing.

Outlier logic means the patterns come from real post-level behavior, not pundit opinion about the category. It means the concept slate is grounded in mechanics that are already breaking out, which is a different document than "five ideas we came up with on a call."

And critically: Virlo metadata is labeled metadata-only until video intelligence actually runs on the video asset. The agent will not infer visual beats from a caption. If you see a claim about a visual mechanic, it's because the video got analyzed, not hallucinated.

Research modes (honest about what you have)

The kit ships with 7 research modes. Each mode has a different connector requirement and produces a different confidence label. The mode label is never faked.

If a connector is missing, the mode label degrades honestly. It doesn't silently pretend. quick-pass means Hypothesis only. outlier-pass means Directional evidence. deep-pass with video intelligence running means Strong directional. Everything the brief claims is tagged at the confidence level the evidence actually supports.

This is the opposite of how every other strategy AI works. Most tools are incentivized to sound certain. This one is incentivized to be honest.

What this agent refuses to do

The non-negotiables baked into every stage. Literally the hard rules:

• Never imply creators are booked, sourced, or committed. Creator names are reference comps only. Not a roster. Not a shopping list. Not "already lined up." Until sourcing has actually happened.

• Never infer visual beats from captions. Virlo metadata is metadata-only until video intelligence runs on the actual asset and tags it as analyzed.

• Never call Virlo Tracking "market research." Comment-derived sentiment is directional signal. Labeled that way in every output.

• Never ship an outlier-pass run as client-ready until synthesis has materialized the final brief. The gate is intact.

• Never fill gaps with vibes. Missing connectors degrade the mode label instead of the agent making stuff up to paper over them.

If you've ever had a creator strategy doc get roasted because a client caught one hallucinated creator claim, these are the rules that stop that from happening.

The anti-positioning block

Section 7 of the brief is the section that changes how clients react to the doc.

It's a list of the 5 generic plays the strategy is explicitly rejecting. Before any concept gets recommended, the brief commits to what it will not do.

Things like:

• "We're not running polished lifestyle montages. The category rewards credible discovery."

• "We're not briefing 'authentic UGC with 3 hook variants.' We're briefing specific hook structures tied to specific objections."

• "We're not picking creators by follower count. We're picking by archetype fit to the breakout pattern."

• "We're not running a generic 'day in the life' format. The outliers in this category are confession-driven, not routine-driven."

• "We're not testing all concepts at once. Three tests, ranked, with decision windows."

When a client reads that section, they stop treating the brief like a deck and start treating it like a plan. The rejection is what makes the rest of the document feel real.

Integrates with the rest of the kit set

This is the tenth kit I've open-sourced in 10 weeks (ok that’s kinda cool, just counted and realized that). Ten weeks, ten kits, all free. And the recent ones are designed to fit together as a loop: the Meta Ads Kit runs your paid distribution, the Landing Page Factory builds angle-matched pages, and the Outcome Kit tells you which angles are actually producing buyers.

Creator Breakout Kit is the layer that sits in front of all three and decides what to test.

  ┌─────────────────────────────────────────────────────┐
  │              THE FULL CAMPAIGN LOOP                   │
  │                                                       │
  │  CREATOR BREAKOUT KIT  → decides WHAT to test        │
  │         │                  (concepts, angles, hooks) │
  │         ▼                                             │
  │  LANDING PAGE FACTORY  → builds the matching         │
  │         │                  landing pages for each    │
  │         │                  angle                     │
  │         ▼                                             │
  │  META ADS KIT          → runs the paid distribution  │
  │         │                  (+ StealAds for creative) │
  │         ▼                                             │
  │  OUTCOME KIT           → tells you which angle is    │
  │                            actually producing buyers │
  │                            (not just clicks)         │
  │                                                       │
  │          THE LOOP  ↺                                 │
  └─────────────────────────────────────────────────────┘

Creator Breakout Kit is the strategy layer before the paid stack. It decides what angles are worth building pages, creative, and test budget around. Then the other kits take it from there.

Before this kit existed, the loop started at "pick some ads to run." Which meant the strategy was retroactive. You'd find out what worked after you'd already burned the test budget on wrong angles.

This one moves the thinking to the front. You decide what to test before you source, before you produce, before you spin up creative. Then every downstream kit is pointed at the right thing.

The cost

  ┌────────────────────────┐     ┌────────────────────────┐
  │  THE OLD WAY            │     │  THE AGENT WAY         │
  │  ──────────             │     │  ──────────            │
  │  Creator strategist:    │     │  Claude Code: included │
  │  $12K/mo                │     │  Firecrawl: free tier  │
  │                         │     │  Virlo: paid, cheap    │
  │  Virlo seat: $500/mo    │     │  Creator Breakout Kit: │
  │                         │     │    free, MIT           │
  │  Research analyst:      │     │                        │
  │  $6K/mo                 │     │  Total additional:     │
  │                         │     │    ~$0 to $100/mo      │
  │  Total: $18,500/mo      │     │                        │
  │  Still vibes            │     │  Result: a real brief  │
  └────────────────────────┘     └────────────────────────┘

quick-pass mode runs with zero connectors. Zero. You still get a brief, just with Hypothesis-only confidence labels instead of Strong directional. Add Firecrawl and Virlo when you want outlier-pass for client-facing work. Add video intelligence when the prospect is high-value enough to justify the paid strategy asset.

MIT license. Fork it. Use it on client work. Use it on your own brand. I'm not charging for the strategy layer.

Getting started

Clone the repo, run preflight, configure your sources. Setup takes an afternoon.

GET CREATOR BREAKOUT KIT (FREE)

What you need:

• Claude Cowork, Claude Code, OpenClaw, Hermes or any agent that reads skills (the kit is 7 markdown skills plus orchestrator scripts)

• Firecrawl API key — free tier works for brand-pass mode

• Virlo API key — required for outlier-pass (the recommended default for client-facing work)

• Optional: ScrapeCreators + OpenRouter for the video intelligence stage (upgrades evidence from metadata-only to Strong directional)

The workflow:

Step 1: Clone the repo and drop in your API keys.

Standard stuff. Clone the repo, copy .env.example to .env, fill in your Firecrawl and Virlo keys (or leave them empty if you want to start in quick-pass mode). Run the preflight check and it'll tell you which research modes are available based on what you've filled in.

Step 2: Tell the agent about your brand.

Open Claude Code (or OpenClaw) in the repo and just talk to it:

The agent seeds the run directory, loads brand context, and asks clarifying questions if anything's missing. Two minutes of conversation per new brand. No CLI flags to memorize.

Step 3: Ask for the outlier research.

Firecrawl pulls brand-level source material from the site. Virlo pulls post-level outliers in your category. All evidence gets attached to the run folder as raw artifacts, not inferred from vibes.

Step 4: Ask for the brief.

The agent runs enrichment, builds the concept slate with the four framework labels, scores every concept by strategic fit and evidence strength, and renders both the markdown version and the dark-terminal HTML artifact. Open the HTML in a browser. That's the show-off version you can send to a partner or paste into a Notion.

Step 5 (optional): Upgrade the evidence.

If the prospect is high-value enough to justify the paid strategy asset, just tell the agent:

It pulls the actual video assets for the strongest outlier posts, analyzes them with Gemini, and re-synthesizes the brief. Confidence labels upgrade from "metadata-only" to "Strong directional." Same brief, much sharper evidence.

You can also ask it pointed questions at any stage:

Who this is for

• Agencies selling creator strategy or managed UGC who are tired of starting every prospect call by making up ideas on the fly

• Brands that want stronger creator briefs before production burns budget

• Paid social teams who need better creator angles, not just more creators

• Operators building a strategy-first creator sourcing pipeline instead of a "send the ideas" reactive one

Not for you if you have a senior creator strategist on payroll who is already running post-level outlier analysis, mapping buyer tension to hook structures, and rejecting generic plays in writing. You don't need this. You are the 1%.

Also not for you if you want the agent to source the creators for you. It does not. That's deliberate. Creator names in the brief are comps only. Sourcing is a separate step, and it should be, because the strategy layer and the sourcing layer are two different jobs.

The honest limitations

V1 is TikTok + Reels focused. The breakout pattern research pulls from short-form platforms. YouTube long-form support is coming but isn't native yet.

Virlo is the paid dependency. The recommended default mode (outlier-pass) requires Virlo (you can now pay for API use). The kit will run without it in quick-pass or brand-pass mode, but the concept slate will be labeled Hypothesis-only. Decide what the output needs before you pay.

Video intelligence is opt-in. Upgrading evidence from metadata-only to Strong directional means actually pulling and analyzing video assets. That has a cost per run. I wired it up so you decide per prospect whether the strategy asset is worth the upgrade.

The brief is a test plan, not production. It tells you what to run first, alongside, and backlog. It does not write the actual creator brief for sourcing. That's the next step, and I wrote a separate sales-handoff skill that helps with follow-up copy but does not replace a real production brief.

Creator names stay comps. Always. The kit will never tell you a creator is booked, available, or committed. That's a guardrail, not a limitation, but it's worth calling out because it's different from tools that imply a roster.

Get the kit

Full system free on GitHub:

You get:

• 7 skills covering the full pipeline from brand truth extraction to client-ready brief

• The 8-section Creator Breakout Brief with executive summary, breakout thesis, pattern board, archetypes, concept slate, first 3 tests, anti-positioning block, and managed next step

• Two render formats from one data source — markdown for editing and agent consumption, dark-terminal HTML for showing a prospect

• Outlier logic via Virlo (post-level breakout ratios, not follower vanity)

• Framework labels on every concept — Breakout Driver, Buyer Tension, Creator Delivery Style, Test Priority

• Confidence-aware evidence labels that never fake certainty (Strong directional / Directional / Hypothesis only)

• The anti-positioning block that forces the brief to commit to what it will not do

• 7 research modes that degrade honestly when connectors are missing

• A sample brief in shared/examples/sample-breakout-brief.html — open it to see exactly what a prospect gets

• Full CLAUDE.md, skill docs, adapter scripts, and evals. Runs on Claude Code, OpenClaw, or any agent that reads skills. MIT license.

No upsell. Fork it.

Stop sourcing creators before you strategize concepts. Start from what's actually breaking out.

Go big,
Matt

P.S. If you missed the recent OpenClaw kits, they're all still free and they plug into each other: Outcome Kit, Landing Page Factory, ScrollClaw, Meta Ads Kit, SEO Kit. This one is the strategy layer that runs before any of them. It decides what angles are worth the other kits' attention in the first place.

We’ve been running these agents for founders. Reply to this email and say ‘agent’ if you want on the list.

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

Your best-performing ad is lying to you.

Right now, the ad your team is celebrating for getting the cheapest clicks is secretly burning through your budget and bringing in zero actual buyers. I call this the 'Fake Winner' trap, and it happens in almost every single ad account I audit.

But last week, I built an automated, 3-part AI system that catches these fake winners instantly.

And today, I'm going to show you exactly how it works, and give it to you for free, so you never waste another dollar on bad traffic again.

Get it free, open-source, below (scroll to the bottom).

Sponsored

Turn AI into Your Income Engine

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

• A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential

• Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background

• Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

The fake winner problem (and why dashboards make it worse)

Most teams don't have a data problem.

They have a fake winner problem.

Here's how it works. You open Ads Manager. You sort by CTR or cost per lead. You see a winner. You scale it. Your team feels productive. Everyone's happy.

Except nobody checked what happened after the click.

The ad with the best CTR often brings the worst buyers. The cheapest lead often books the fewest calls. The campaign report that looks great in-platform hides the weakest downstream outcomes.

I've been running paid media for clients for years. This pattern shows up in almost every account I audit:

Meta optimizes for Meta. Their reporting rewards engagement. Your business needs buyers.

And here's what makes it worse: most businesses don't have pristine attribution. No clean CRM. Broken UTMs. Calendly bookings in one system, ad spend in another, landing page data somewhere else.

So they look at what they can see (platform metrics) and make decisions. And those decisions are often exactly wrong.

I asked a founder I’m working with last month: "Which of your ad angles is producing the best customers?" Not clicks. ACTUAL paying customers.

Dead silence. They had a dozen dashboards open and couldn't answer the question.

What I built

Outcome Kit is a multi-agent system that reads your ad data, your landing page data, and whatever business outcome signal you actually have (leads, bookings, signups, purchases, revenue) and tells you four things every morning.

It's built for Cowork, the Claude Code multi-agent framework. Three agents coordinate the pipeline. Each one has a clear job and does NOT do the other agents' work.

Also works with OpenClaw and Hermes if that's your stack.

Let me walk you through the four outcome truths.

The Four Outcome Truths

Every run classifies your funnel into four buckets. This is the framework.

1. Real Winner

Strong upstream response AND strong downstream outcome quality. The ad angle that drives clicks and drives buyers. This is what you scale.

2. Fake Winner

Looks good in-platform. Weak on actual outcomes. High CTR, cheap leads, garbage conversion downstream. This is what you think is working. It's burning your budget.

3. Leak

The interest is real, but something after the click is breaking. Strong ad performance, weak page conversion. Message-page mismatch. The ad promised one thing, the landing page delivered something else. The fix isn't the ad. It's the page (or the follow-through).

4. Underfed Winner

Low volume, strong efficiency. The angle that's quietly outperforming everything else but only getting 9% of your budget. This is the one you're starving while you feed the fake winner.

That's the entire diagnosis model.

Why this works when most "attribution" tools don't

Most attribution tools assume a world that doesn't exist.

They want pristine UTMs. Perfect CRM hygiene. Fully connected revenue data. A data engineering team maintaining the pipes.

That's not reality for 95% of operators.

Outcome Kit assumes the real world:

The key: every recommendation comes with a confidence tag.

The agent doesn't fake certainty. It tells you how sure it is:

• High confidence: "this angle is your strongest by cost per booked call, stable across 14 and 30 day windows". Act on this now

• Medium confidence: "this creative family is probably bringing low-intent traffic, but sample is still small". Watch it, don't kill it yet

• Low confidence: "revenue linkage is too weak to call this". Recommendation withheld

When's the last time your dashboard said "I don't have enough data to tell you"? Never. It just showed you a number and let you make a bad call. This agent stack would rather say nothing than say something wrong.

Why I stopped looking at “campaigns”

Here's the moment that changed how I think about this.

I was in a new ad account last quarter. They had 6 campaigns, 18 ad sets, 40+ ads. I asked their team a simple question: "Which message is actually producing buyers?"

Nobody could answer. Because the data was organized by campaigns, not by message. The same "competitive teardown" angle was scattered across three campaigns and two audiences. The "founder-led proof" message was split between a prospecting campaign and a retargeting campaign. There was no way to see what the angle was doing across the whole account.

That's the wrong cut.

"Ad 27" is not the strategic question. The strategic question is: is the competitive teardown message producing stronger buyers than the time-savings message?

Outcome Kit is built around angles. Not campaigns. Not ad sets. Message families.

When the agent says "the competitive teardown angle is your underfed winner," you know exactly what kind of creative to make more of. When it says "founder-led psychology ads are leaking on the generic SaaS page," you know the ad is fine and the page needs work.

That's how we get to "your best message is starving while your worst message eats your budget."

The 12 skills

Each skill runs standalone or as part of the full pipeline.

Three agents, one pipeline

This is the part that's different from every other kit I've given away.

Outcome Kit doesn't use a single agent running 12 skills sequentially. It uses three specialized agents that coordinate through shared artifacts. Each agent has explicit guardrails about what it owns and what it doesn't touch.

Agent 1: Data Reader. Runs Meta source reader, page signal reader, and outcome event reader. Validates that the raw data exists and isn't empty. Surfaces trust notes and quality warnings. Does NOT interpret strategy. Does NOT declare winners. Its only job is source truth.

Agent 2: Diagnostician. Maps angles, scores them against your primary outcome, and classifies everything into the Four Outcome Truths. This is the judgment layer. It decides what's real, what's fake, what's leaking, and what's underfed. Categories are mutually exclusive. It flags confidence concerns but does NOT write recommendations.

Agent 3: Brief Writer. Takes the diagnosis and turns it into concrete operator moves (scale, cut, rewrite, rebuild, test). Writes the daily brief. Persists patterns to memory so the system learns over time. The brief must be scannable in under 60 seconds.

They hand off through artifact files in workspace/outcomes/. Agent 1 writes raw data. Agent 2 reads it and writes scored diagnosis. Agent 3 reads that and writes the brief.

This matters because each agent stays in its lane. The data agent never jumps to conclusions. The diagnostician never softens its findings to make the brief sound nicer. The brief writer never reinterprets the data.

If you've ever used ChatGPT to "analyze my ads" and gotten a wall of hedge-everything mush, this is the fix. Separation of concerns. For AI agents.

The repo also ships with pipeline-order hooks (a shell script that blocks later agents from running on missing or invalid data) and confidence rules that actively suppress recommendations when evidence is too weak. The agents literally cannot skip steps or overclaim.

What this agent refuses to do

Here's what I put in the system's soul file. Literally the first lines:

Most dashboards and AI tools do all three of those things by default.

I built this kit to explicitly defend against the dumb stuff:

• Declaring winners from CTR alone. The number one sin of paid media reporting.

• Treating cheap leads as high-quality outcomes. A $5 lead that never books a call costs you infinity dollars per customer.

• Hiding broken tracking instead of surfacing it. If the UTMs are garbage, the agent says so instead of pretending the data is clean.

• Flattening different message angles into one blended average. Your "time-savings" angle and your "competitive teardown" angle are not the same thing. Averaging them together tells you nothing.

• Reporting numbers without decisions. If the output doesn't end with "do this next," it's dashboard theater. And we've all sat through enough of that.

• Overclaiming certainty when confidence is low. When the data isn't there, the agent withholds the recommendation entirely. It doesn't guess. Most tools guess and dress it up with a chart.

If you're already running my Meta Ads Kit

The Meta Ads Kit blew up when I dropped it. 634K views on X. Thousands of clones.

Twitter tweet

And it's great at what it does: pause bleeders, scale winners, shift budget, catch creative fatigue.

It's the operator layer. It watches what's running and acts on the platform signals.

But here's the gap: the Meta Ads Kit optimizes for platform metrics. It'll scale an ad with strong CTR and low CPA. That's its job.

The problem is that sometimes the ad with the best platform metrics is your worst performer downstream. The Meta Ads Kit would happily scale your fake winner because Meta says it's winning.

That's what Outcome Kit fixes.

Run both. The Meta Ads Kit handles the daily operations. Outcome Kit tells you whether those operations are pointed at the right things.

The Meta Ads Kit is the hands. Outcome Kit is the eyes.

And if you're using StealAds for creative generation, this closes the whole loop:

Before Outcome Kit, the creative loop had a blind spot. StealAds would generate more variations of whatever the Meta Ads Kit said was winning. But if that "winner" was a fake winner (high CTR, garbage buyers), you'd be generating more creative for the wrong angle. More ads that look good in Ads Manager and fail downstream.

Now the signal is clean. Outcome Kit identifies the real winner angles. StealAds generates creative for those. Meta Ads Kit runs them. Outcome Kit checks the results. The loop tightens every cycle.

Getting started

Clone the repo, run the doctor check, configure your sources. Setup takes one afternoon.

GET OUTCOME KIT (FREE)

What you need:

• Meta Ads access token and ad account ID (your existing setup works)

• GA4 property ID + service account JSON (for page signal data)

• One outcome source: Calendly API token, HubSpot private app token, or a simple CSV/JSON file

• Cowork, OpenClaw, or Hermes installed

The workflow:

Step 1: Clone and configure.

git clone https://github.com/TheMattBerman/outcome-kit.git
cd outcome-kit
cp .env.example .env
cp config.example.json config.json

Add your API keys and pick your primary outcome (bookings, signups, purchases, whatever matters to your business).

Step 2: Define your angles.

This is the most important 10 minutes. Map your message angles to creative families and landing pages. Start simple:

[
  {
    "angle": "time-savings",
    "audience": "heads-of-growth",
    "creative_family": "founder-direct",
    "page": "/time"
  },
  {
    "angle": "competitive-teardown",
    "audience": "founders",
    "creative_family": "ugc-analysis",
    "page": "/teardown"
  }
]

You can also add source enrichment rules so the system auto-tags outcomes from Calendly or HubSpot by matching naming patterns to angles.

Step 3: Run the doctor.

npm run doctor

This validates your config, checks your API connections, confirms your angle mapping, and tells you if anything is missing before you run the real thing.

Step 4: Run the sample pipeline.

npm run run:sample

See the output format with sample data before you connect your real accounts.

Step 5: Run for real.

npm run run

The repo ships with full CoWork support out of the box. That means CLAUDE.md, three agents in .claude/agents/, skills in .claude/skills/, rules, and hooks. If you're already using Claude Code or Cowork, the agents pick up the pipeline structure automatically.

You can also talk to it naturally: "Find fake winners in my Meta funnel." "Which angle is driving booked calls?" "Give me the daily outcome brief."

Works with OpenClaw and Hermes too.

Step 6: Set up a cron.

One cron job. The three agents coordinate the full pipeline. The brief lands in Telegram, Slack, or wherever you want it.

You start your day knowing what's real.

The cost

Every API this connects to has a free tier or is already something you're paying for. The three agents, the 12 skills, the pipeline, all open source. MIT license.

Who this is for

• You're spending money on paid traffic and you're not sure which message angle actually produces buyers (not just clicks)

• You run an agency managing multiple accounts and you need outcome-level truth, not platform-level vanity

• You're a SaaS founder with demo bookings or signups as your core metric and you're tired of sorting by CPL

• You have messy tracking, disconnected tools, and still need to make decisions anyway

• You're an e-commerce brand that knows CTR and ROAS tell different stories depending on the creative

Not for you if you have a 5-person data team, pristine multi-touch attribution, and a Snowflake warehouse already doing this. You don't need this. You're the 5%.

Also not for you if you want one-click magic. You still need to define your angles, configure your outcome source, and read the brief. The agent does the analysis. You make the calls.

v1 limitations

V1 is Meta-first. Google Ads support is coming but isn't native yet. If Meta is your primary paid channel, you're good. If you're Google-only, wait for V2.

You need at least one outcome source. A Calendly token, a HubSpot connection, or even a CSV file of your bookings/purchases. If you have zero downstream data, the agent can't find fake winners because it has nothing to compare against.

Angle tagging is manual to start. You define the angle map. The system doesn't auto-discover your message strategy. That's actually a feature (forces you to think about it), but it's 10-15 minutes of work upfront.

Confidence scales with data. Week 1 with 30 conversions? You'll get medium-confidence reads. Week 4 with 200 conversions? High confidence. The system is honest about this and will withhold recommendations when the data isn't there. That's by design.

No automatic budget changes in V1. The agent recommends. You approve. Same philosophy as my Meta Ads kit. Start with visibility. Graduate to autonomy when you trust it.

Get the kit

Full system free on GitHub:

You get:

• 3 specialized agents (data reader, diagnostician, brief writer) that coordinate through shared artifacts

• 12 skills covering the full pipeline from data ingestion to daily brief

• Full CoWork support out of the box (CLAUDE.md, .claude/agents/, skills, rules, hooks). Also works with OpenClaw and Hermes.

• Angle-first analysis that cuts by message strategy, not campaign structure

• Confidence-aware scoring that never fakes certainty

• The Four Outcome Truths framework (real winner, fake winner, leak, underfed winner)

• Pattern memory that compounds across runs

• Doctor script that validates everything before you run

• Setup docs, config templates, sample data, the works

No upsell. MIT license. Fork it.

Stop trusting vanity metrics. Start acting on what's actually producing outcomes.

Go big,
Matt

P.S. If you missed the last few OpenClaw kits, they're all still free: SEO Kit, Meta Ads Kit, First 1000 Customers, UGC Factory, ScrollClaw. This one plugs into the Meta Ads Kit if you're already running it. Same agent, new superpower.

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

You’re running 12 different ad angles. You have different hooks, different audience segments, and different promises. But when someone clicks...

...you send them all to the exact same landing page.

The click dies right here.

But what if you could make a custom, high-converting landing page for every single ad angle in just 12 minutes?

I'm going to show you how to do exactly that with Claude Cowork.

And to prove it, I took my own SaaS, StealAds, and built six completely different, deployable landing pages in an hour.

And I’m giving the whole kit away for free (and open source). Scroll and get it below.

Let's build.

Sponsored

How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more

The real problem nobody’s talking about

The real problem nobody in marketing talks about is the gap.

Twitter tweet

If your ad talks targets time saving, saying "stop wasting 8 hours on… {whatever} " But your landing page talks about health, saying, “this {product} helps improve your {whatever}” then you’ve introduced friction in your audience’s mind.

Your ad might target founders, but your lander talks to everyone. Founders, freelancers, agencies, all at once.

Your ad makes one promise. Your page makes a different one.

Intuitively you know this, and you’re not lazy. But building a good landing page takes two to four weeks of copy, design, revisions, and QA. By the time the page is live, your ad strategy has already moved on.

The bottleneck was your production speed.

Claude Cowork + Landing Page Factory

I built seven skills that Cowork loads and runs in sequence. Paste a URL. Tell it the angle. It handles the rest.

If you don’t know, Cowork is a desktop agent from Claude/Anthropic. You don’t need your terminal, you don’t need a Mac mini. Download the app and roll.

Twitter tweet

Here's what happens when I point my landing page factory at StealAds:

A complete landing page.

The neon green buttons, the hot pink accents, the terminal aesthetic with sharp corners and JetBrains Mono. This is all pulled automatically from our live site. The copy explains why StealAds works (psychology decoding), not just that it exists. The design doesn't look like AI made it.

12 minutes from URL to deployable page.

Why this works and "make me a landing page" doesn't

Here's what happens when you ask any AI to make a landing page.

It generates a headline like "Unlock Your Business Potential." It picks some blue and purple gradient. It throws in three feature cards with vague descriptions. It invents a testimonial from "Sarah M., Marketing Director" who definitely doesn’t exist. And you get a page that looks exactly like every other A slop page on the web.

AI can be amazing. But the AI doesn’t know anything real, and it doesn’t know anything about you!

My landing page factory kit has seven stages. And each one exists to prevent a specific failure mode.

Stage 1: Extract

Give it a URL. It scrapes everything that matters about the brand.

It scrapes your actual site with Firecrawl.

Not just the text. The actual visual identity. Colors as hex codes. Font families. Spacing system. Logo. Trust badges. Pricing. Every CTA. Every claim. Every testimonial.

Under the hood, this is Firecrawl's v2 branding format. One API pulls the entire brand DNA, colors, fonts, component styles, dark/light scheme. A second pass grabs the full page as markdown for claim extraction.

Here's what's different from "paste a URL into ChatGPT": every claim gets a tag with where it came from.

• Did the site say "500K users"? That's verified_source.

• Did we infer they target enterprise? That's derived_source.

• Can't find proof for a claim? It gets flagged as missing.

Stage 2: Strategy

This is the stage no AI page builder has. It's also the reason every AI landing page looks the same.

The strategy layer takes the raw info (that we got above) and makes it decisions.

• What's the unique mechanism?

• What claims are we allowed to make?

• What proof is real vs. placeholder?

• What kind of page should this be?

The strategy decides which route to take. Not the model's taste. Not a template picker. Actual routing logic.

Hard rule: No strategy, no page. 

Stage 3: Brand Profile

This builds a two-layer system.

Layer 1 is only observed evidence. These are exact phrases the brand uses. Visual identity it actually saw. Claim patterns from the site.

Layer 2 is synthesized. This is the model figuring out positioning, audience model, voice dimensions, and language bank.

When the agent writes copy later, it knows if a phrase came from the brand itself mouth or from Claude.

Stage 4: Copy

The hard ban list kills AI slop at the source.

Stage 5: Page Visuals

Every image gets classified before anyone writes a prompt.

Stage 6: Page Build

The taste layer. A strict hierarchy that prevents AI defaults.

Stage 7: Page QA

Nothing ships without passing the gate.

Six pages, six angles, one hour

One page is a start. Six pages for six angles is a system.

Here's what I built for StealAds:

Same product. Six audiences. Six emotional architectures. Same neon green and hot pink brand palette. Same terminal aesthetic. Same JetBrains Mono. But completely different persuasion logic.

Each page preserves the mechanism (psychology decoding) and frames it through the lens that matters most to that specific buyer.

If I'm running Meta ads with a time-savings hook targeting Heads of Growth, they land on the Time page. If I'm running a competitive intelligence angle targeting founders, they land on the Competitive page.

The ad promise matches the page. Every time.

The matrix

When building a page takes 12 minutes instead of 3 weeks:

Start with your top 2 audiences and top 2 angles. That's 4 pages. Test them. Let the data tell you which emotional architecture converts for which buyer.

Then build more. Because now you can.

The full pipeline

Take a step back and look at what we've built over the last few months.

The Meta Ads Kit runs your Meta ads. The Google Ads Copilot manages your Google ads. StealAds makes your creative by decoding the psychology behind winning ads and generating branded versions for you.

And now Landing Page Factory builds the pages those ads point to.

That's the whole pipeline. Creative. Ads. Pages. Connected.

Get the kit

Seven skills for Claude Cowork. Also works with OpenClaw, Hermes, Cursor, or any agent that reads markdown.

Claude Cowork (easiest): Upload the skills/ and references/ folders to a Cowork project. Start chatting. Say "Build me a landing page for [URL]." That's it.

OpenClaw: Clone, install, run. Same pipeline, plus cron automation and multi-agent orchestration (you know I love my Openclaw).

No catch. No upsell. MIT license.

If you want the full stack (Meta Ads Kit for ad angles, Google Ads Copilot for search intent, Landing Page Factory for matching pages) all three are free on GitHub. They're built to work together.

If you'd rather have us run this for your brand, that's what we do at Emerald Digital. Reply and we'll talk.

Matt

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

We’ve all been lied to about UGC.

We’re told you need real creators, paying $800 a pop to get conversion.

But that campaign you just saw rip across the feeds? It isn’t real.

UGC creators can be a headache. But most AI UGC looks fake.

The moment your brain sees "perfect," it scrolls. The real secret isn't the AI. It’s mining exact phrases of human frustration from Reddit, then forcing the AI into an "anti-polish" doctrine.

That best-performing video wasn't born in a UGC campaign.

I built this OpenClaw skill to mine 1-star reviews and generate this viral, realistic content in under an hour.

Get it free, open-source, below (scroll to the bottom).

Sponsored

How Marketers Are Scaling With AI in 2026

61% of marketers say this is the biggest marketing shift in decades.

Get the data and trends shaping growth in 2026 with this groundbreaking state of marketing report.

Inside you’ll discover:

• Results from over 1,500 marketers centered around results, goals and priorities in the age of AI

• Stand out content and growth trends in a world full of noise

• How to scale with AI without losing humanity

• Where to invest for the best return in 2026

Download your 2026 state of marketing report today.

Get Your Report

Why AI UGC fails (and it's not what you think)

Most people trying to make AI UGC hit the same wall.

They prompt Sora or Kling. They get something. The person is gorgeous. The lighting is perfect.

And nobody believes it for a second.

The problem is that it looks produced. And produced is the enemy of UGC.

UGC works for one reason: it looks like a real person pulled out their phone and yapped. That's it. The moment it looks like anything else (an ad, a brand video, a product commercial) your brain sees it as marketing and moves on.

I call this the anti-polish doctrine:

Every decision in this pipeline routes through one question: could a real person have shot this on their phone?

If the answer is no, fix it. If you can't tell, that's a pass.

The moment I knew the script was the problem

The second failure is more subtle than the visuals. And it costs you more.

In January, I ran a UGC test for a DTC client with four UGC vids.

We used the same creator, it was for the same product, and they had the same vibe. And one of the videos outperformed the others by 3x conversions.

The winning script had one thing the others didn't: we stole a line directly from a 1-star Amazon review.

"I was doing everything right and my body wasn't cooperating."

No copywriter would probably write that. It sounds too defeated. But someone who had actually experienced that, maybe wrote it on Amazon at 1am, that person felt it. Viscerally.

That's the key.

Specific language helps viewers imagine the result. And generic language doesn’t. The first converts. The second doesn't.

This is why this skill starts with persona research. Sure, you can pull competitor reviews on Amazon, TrustPilot or Reddit. Or, better yet, ask your Openclaw to do it. You aren’t pulling information. You want to pull exact phrases.

You're looking for:

• "I was doing everything right and my body wasn't cooperating" → goes into a script

• "I finally feel like myself again" → useless, discard

• "I wore a dress I hadn't put on since my daughter was born and cried in the fitting room" → this is what you want

What I built

The sora-ugc skill is an OpenClaw pipeline that takes a brand brief and outputs finished UGC-style video campaigns.

Let me walk you through each piece.

The first-frame gate

Before we get to video, let's talk about the snap decision that happens the first 3 seconds.

Your viewer's brain decides to stay-or-scroll even before they are conscious of why.

By the time they've read your hook (if you’re that lucky), that decision is already made. Frame 1 is a visual design problem.

This is why the pipeline generates frame 1 separately, with Nano Banana 2, before animating anything with Sora.

Here’s the prompt that makes it work:

Raw iPhone 15 Pro photo. Candid moment, unfiltered, authentic.
f/1.8 aperture, shallow depth of field, slight digital sensor grain.

REALISM RULES:
- Natural imperfections: visible skin texture and pores, flyaway hairs
- Under-eye area shows natural shadows, not concealed
- Environment has real clutter, not styled
- Lighting is practical (window, lamp, overhead) - never ring light
- Expression is candid mid-moment, not posed for camera
- Off-center framing, phone propped, not professional

Negative prompt: CGI, perfect skin, beauty filter, studio lighting,
ring light, retouched, airbrushed, stock photo, model pose

This block goes at the top of every frame 1 prompt. Before the color JSON. Before the scene description.

The specific AI ‘tells’ this fights:

• Skin smoothing. AI removes pores. Counter it: "visible skin texture and pores."

• Symmetrical faces. AI makes faces too symmetrical. Counter it: "slight facial asymmetry."

• Grey color cast. Nano Banana defaults to high contrast images. Fix it with JSON color extraction from a reference image that matches your environment.

• Clean environments. AI backgrounds are spotless. Counter it with specific clutter items by name: coffee mug, phone charger, hair tie. Not "messy." Specific objects.

• Perfect composition. AI centers and balances. Counter it: "off-center, phone propped on stack of books, not professional."

You might make 2-4 frame attempts before one passes. But that iteration is worth it, the first frame locks the visual world for everything that comes next.

Two engines, one video

Most people pick one AI video model and deal with its limits. But this pipeline uses two:

Sora 2 handles A-roll: the talking head. This is where Sora 2's lip sync is genuinely good. It makes confessional, casual, real-person energy. Frame 1 into Sora i2v, structured motion prompt, download immediately.

Twitter tweet

Kling 3 handles B-roll: everything that isn't the creator talking to camera. Product close-ups. Product in use. Establishing shots. Before/after scenes. You get it.

The same character reference image feeds both engines.

Same face in the Sora talking head. Same face in the Kling B-roll. Consistent creator across every clip in the video.

I use Kling for speed and the Element system. You can use (@Element1 for face, @Element2 for product) to lock both a specific character AND a specific product into every B-roll shot.

And the script gets marked before you generate anything:

[A-ROLL: Sora - talking head, car, hook]
"so I'm sitting in my car right now… and I know that sounds
weird but like… this is the first time in three years I'm
leaving at a normal hour."

[B-ROLL: Kling - back office, billing screen]
(Show: cluttered back office, laptop open with billing dashboard,
@Element1 hunched over keyboard, late at night, overhead
fluorescent light.)

[A-ROLL: Sora - talking head, car, close]
"anyway. I'm going home. that's - yeah. that's the whole thing."

A-roll drives the timeline. B-roll swaps visuals only. Then the audio pipeline brings it all home.

The voice consistency problem (and the fix)

Here's what breaks multi-clip UGC:

Every Sora clip has its own voice. But those voices can vary wildly between clips. Stitch three clips together and it sounds like James McAvoy in Glass.

Sora does not let you create characters that look like humans, and I don’t know about you, but most of my videos need humans in them.

The solution is ElevenLabs speech-to-speech, and the order you run it matters:

A-roll drives the timeline. If you stitch A+B first then try to sync audio, the timing breaks because B-roll adds visual duration that doesn't exist in the voice track.

The six formats

One skill. Six ways to use it. Each format exists because it produces a distinct behavior shift the others can't.

1. Talking Head Yapper - The workhorse. One person, one selfie angle, honest opinion. Phone propped, natural light, specific environment (bedroom, car, bathroom counter). 15-25 seconds, 2-3 Sora segments. Where most campaigns should start.

2. Hook Face + Demo - Classic format that actually works in high-volume UGC right now. Zero dialogue in the hook. Just a face with an emotion (shocked, frustrated, crying, happy, confused, angry, six emotion categories with specific Sora prompts for each) and text overlay. 2-5 seconds. Hard cut to the demo. The hook face catches attention at the visual processing level before the conscious brain engages. The text gives the brain a reason to stay. The demo delivers the content.

3. Wall of Text - Overwhelmingly long on-screen text that creates a curiosity trap. 4-5 seconds of dense text - more than can be read in the clip duration. The viewer's brain starts reading and needs to finish (so they watch on a loop). This beats the scroll reflex by hijacking the completion impulse instead.

4. Visual Transformation Story - Named concept hook ("the divorce effect"), 3-5 image segments animated with Sora. The NAME is the scroll-stopper. "The divorce effect" stops the scroll. "How divorce changed me" doesn't.

5. Hybrid Transformation - This is the format that crossed $100k in a recent campaign. Talking head bookends with a voiceover slideshow mechanism bridge in the middle. The talking head creates identification. The slideshow maintains attention through the mechanism section where pure talking-head retention drops. The talking head close delivers the transformation moment with eye contact. For anything where "why it works" is necessary for conversion.

6. Podcast Clip - Fake podcast. Shure SM7B aesthetic, over-ear headphones, warm amber lighting, wood slat backdrop. Creator as "guest," mid-conversation. Viewers perceive podcast guests as credible experts, not advertisers. Authority from context, not claims. One note on prompting this: describe how the audio sounds, not what gear produced it. "Clean, natural podcast audio, close and present with subtle room tone" works. "Sounds like a Shure SM7B" does not.

The virality gate

One update that I wish I had on the first round: mandatory virality scoring before anything gets published.

Seven criteria, each scored 0-100:

Hard gate: if hook strength < 50, automatic NO-GO. Doesn't matter what the other scores are.

Nothing else matters if they scroll past in the first two seconds.

The scoring uses a vision model (Gemini Flash, Claude) to evaluate either the full video or five evenly-spaced frames. It returns a weighted score, a GO/NO-GO, and, importantly, one specific, implementable fix with estimated score improvement.

Getting started

Download the skill from GitHub, run the dependency checker (it verifies all your API connections before you touch a prompt), and install. The whole setup takes 10 minutes.

GET SCROLLCLAW (FREE)

What you need:

• fal.ai key — Sora 2 for A-roll, Kling 3 for B-roll

• OR Replicate key — Nano Banana for first frame generation

• OpenRouter key — Gemini for virality scoring and video analysis

• ElevenLabs key — optional, only needed for multi-clip formats where voice consistency across segments matters

• OpenClaw installed

The workflow:

Step 1: Persona research. Tell your agent to mine Amazon reviews, Trustpilot, Reddit for your product category. It comes back with exact phrases, not paraphrases. You can also do this manually in about 60 minutes. Either way, this step is the whole thing.

Step 2: Create your creator profile.
Tell the skill: "Create a creator profile for Coach Dan. mid-40s, male, slightly weathered look, wearing a polo. Gym environment." One profile per creator. Locked identity, same face, same build, same distinguishing features across every clip, every campaign.

Step 3: Pick your format. Read references/format-library.md. Let the persona research inform the choice. Map your script to the shot breakdown BEFORE writing a word. Every segment should be accounted for.

Step 4: Script it.
Use exact phrases from your persona mining. Test every line. Remember if it sounds like a brand wrote it, it's wrong. Read it out loud. If you'd never say it to a friend, rewrite it.

Step 5: Generate frame 1.
Tell the skill your scene, your creator profile, and your environment. The photorealism pre-prompt fires automatically. You just describe the moment. "Coach Dan in his office, late afternoon light, phone propped on desk, looking directly at camera with a tired-but-relieved expression." Generate context-aware frames per setting. Iterate here more than anywhere else.

Step 6: Animate A-roll.
Point the skill at your first frame and give it the motion prompt. Describe the micro-movements you want. "Minimal movement. Slight natural breathing. Small shoulder shift at the 4-second mark. Phone-propped stillness." The skill sends it to Sora via fal.ai and downloads the clip.

Step 7: B-roll via Kling.

The skill automatically pulls an environment frame from your A-roll and hands it to Kling as the starting image. Then you give Kling the B-roll description: "Cluttered back office, laptop open with a billing dashboard, overhead fluorescent light, late at night." Same visual world. Same floor. Same lighting.

Step 8: Audio pipeline (multi-clip).
Tell the skill to run the audio pipeline on your campaign folder. It stitches the A-roll, runs ElevenLabs S2S for voice unification, inserts B-roll, normalizes resolution, and assembles the final video. For single talking head clips: skip this step entirely, Sora's audio is already right.

Step 9: Post-production.

Apply color grade, grain (4K upscale trick), frame rate lock. If mixing Sora + Kling clips, grade Kling to match Sora's color world — they produce different color profiles by default.

Step 10: Captions — ALWAYS LAST.

Post-production first, captions last. Grain and color grade degrade clean caption pills if you apply them in the wrong order.

Tell the skill your caption lines and it auto-detects your video resolution and builds the overlay. Keep text in the green zone (TikTok/Reels block the top 210px, the right 120px, and the bottom 440px). Platform UI lives there. The skill knows where the safe zone is.

Step 11: Virality score before you publish.

Hook < 50 = don't post. Fix the hook and regenerate.

Step 12: Batch campaigns.
Tell the skill to run your full creator × format matrix: "Run the gym client campaign with all three creators across talking head, hook-demo, and hybrid transformation formats." One brief. Multiple creators. Multiple formats. The skill queues and runs the whole batch.

Who this is for

• You're running paid social and testing UGC angles and want speed

• You manage DTC or app brands that need high-volume creative

• You run an agency and want to productize video creative that doesn't look like AI slop

• You have a brand with a real customer journey

Not for you if you want one-click UGC with no research or ‘taste’. The pipeline makes production fast. The insight and taste makes it convert.

Get the skill

Full system free on GitHub:

Works across DTC, B2B SaaS, and consumer apps. Different categories, different personas, same flow.

No upsell. MIT license. Fork it.

Show me what you’re scaling. I read every reply.

Go big,
Matt

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

Last Tuesday I pointed this Openclaw agent at a Google ads account I was asked to audit. 4,300 search terms. $23,000 in spend over 90 days.

First thing it flagged wasn't waste like I thought. It was a rotten CPA.

The account showed $179 cost per acquisition. Respectable. The kind of number you'd look at in a dashboard and think, okay, we're fine. But the agent broke it down by intent class, and the real story was ugly:

$4,200 a month going to "what is crm" and "crm jobs near me" and "free crm template download." 847 queries. Zero conversions. Just... gone.

A few weeks ago I released my Openclaw agent that runs your Meta ads. And I kept getting one question… “do you have one of these for Google ads?”

Now I do.

I built an Openclaw agent that runs your Google ads. Fifteen skills. Full intent mapping. Daily operator reviews. And I'm giving the whole thing away free (scroll to the bottom for the link).

Here's how it works.

Sponsored

Here’s how I use Attio to run my day.

Attio is the AI CRM with conversational AI built directly into your workspace. Every morning, Ask Attio handles my prep:

• Surfaces insights from calls and conversations across my entire CRM

• Update records and create tasks without manual entry

• Answers questions about deals, accounts, and customer signals that used to take hours to find

All in seconds. No searching, no switching tabs, no manual updates.

Ready to scale faster?

Start now →

What the agent catches (and what you'd never see)

Every account I tested had a version of the same problems. Here's what the agent finds:

1. It finds the fake CPA.

When buyer queries, comparison queries, and educational queries all live in the same campaign, Google reports one blended CPA.

That number is meaningless.

The agent breaks it apart by intent class.

Every account I tested had a fake CPA. Every single one.

2. It catches polluted tracking.

Search partner traffic eating 12% of spend with 0% of conversions.

A 72-hour conversion lag that makes yesterday's data useless.

Double-counted form submissions.

None of this shows up in the Google Ads dashboard.

The agent goes looking every morning.

3. It maps mixed intent.

This is the big one. Someone searches "clio vs hubspot crm." That's comparison intent. They need different copy, a different landing page, a different CPA target than someone searching "crm demo for law firms."

But in most accounts? Same campaign. Same ad group. Same ads. Same landing page.

Google's algo sees both and optimizes for the average. Both experiences are wrong. Both waste money.

Most Google Ads problems aren't bid problems. They're intent-mixing problems. The agent reads 4,000 search terms and classifies every one by intent in 40 minutes.

4. It exposes broken landing paths.

Once you see the intent map, you see the landing page problem immediately.

Comparison shoppers hitting your homepage. Buyers hitting a blog post. Branded searches going to a generic page.

The search term told you exactly what the person wanted. The account structure ignored it.

5. It acts on what it finds.

This is the part that makes it different from every audit tool I've used.

The agent doesn't just tell you what's wrong.

It drafts specific fixes (negatives with match type and scope, structure changes, budget moves) and stages them for your review.

You approve. It applies. Dry run first. Full audit trail. Instant undo.

What the agent looks like

Telegram message from the agent:

I read that in two minutes. Approve the negatives. Greenlight the comparison split. Move on.

Total time managing Google Ads that day: under 5 minutes.

The agent didn't give just recap the numbers. It gave me decisions to act on.

The Intent Map

Every time the agent reviews your search terms, it builds something I call the Intent Map. A living model of what your Google Ads account is actually buying.

The map lives in your workspace. It persists across sessions.

Most PPC tools give you a snapshot. This gives you memory.

"What is [industry term]" is almost always waste. "[Role] jobs near me" is always waste. "Free [product category]" converts at near-zero everywhere.

Over time, the agent builds the same pattern library a senior strategist carries in their head. The difference is it doesn't forget.

How the pieces fit

Fifteen skills. Three layers.

READ  ──▶  DRAFT  ──▶  APPLY

Read layer pulls live account data (search terms, keywords, conversions, budgets) through Google's official MCP server. Or you paste a CSV export. Same analytical engine either way.

Draft layer turns findings into concrete proposals. Not just a "your CPA is high."

These are specific negatives with match type and scope.

Structure changes with before/after projections.

RSA directions pulled from the language real buyers actually use. Every proposal staged in workspace/ads/drafts/ for your review.

Apply layer executes what you approve. Dry run shows you exactly what will change. You confirm. It applies.

Full audit trail logged. Instant undo with /google-ads undo if anything looks wrong.

The difference between this and a thermometer: a thermometer tells you the temperature. This tells you why you have a fever and hands you the medicine.

The truth about Google's AI

Google wants you to trust their automation. Broad match. Smart Bidding. Performance Max. "Let the algorithm do its thing."

Twitter tweet

Smart Bidding is genuinely good at auction-time decisions. But Google's AI optimizes for Google.

They push broad match because it expands inventory. They push Performance Max because it lets Google decide where your money goes. They changed the search terms report to hide "low volume" queries you're paying for but can't see. They build recommendations that auto-apply if you don't manually opt out. Every update in the last three years has reduced advertiser visibility and control while asking for more trust and more budget.

When Google says "let our AI manage your ads," ask: manage them toward what outcome?

Twitter tweet

The winning move: use both. Let Google handle the auction. Use your own agent to manage the strategy layer on top. Read the search terms. Build the intent map. Cut the waste. Shape the structure.

You keep the judgment. They run the auction.

Getting started

Step 1: Get the kit

git clone https://github.com/TheMattBerman/google-ads-copilot.git
cd google-ads-copilot
./install.sh auto

Fifteen skills. Reference docs. Workspace templates. Apply layer with audit trail. Everything.

Step 2: Connect your account

Two options. Connected mode hooks up Google's MCP server and the agent pulls live data automatically. Export mode: paste a CSV from Google Ads. Same analytical engine, same quality. Connected is better for daily ops. Export is fine for a first audit.

/google-ads connect setup

The agent discovers your accounts, picks one, writes the workspace config. If you're using export mode, skip this and go straight to Step 3.

Step 3: Run your first Query Interview

/google-ads search-terms

Drop in your search terms report (or let the MCP pull it live). The agent reads every query, classifies intent, finds waste, and builds your first Intent Map. This is the moment. 4,000 rows of search term chaos turned into a clear intent model with specific cut/isolate/scale recommendations.

Step 4: Run the daily operator

/google-ads daily

The morning review. Set it on a cron schedule. The brief lands in Telegram or wherever you want it.

Step 5: Review, approve, apply

The agent drafts. You review. Approve what makes sense. The apply layer dry-runs the change, you confirm, it executes. Full audit trail. /google-ads undo if you change your mind.

Start with negatives. It’s lowest risk, highest immediate impact. Graduate to structure changes when you trust the reasoning.

The difference is the difference between a thermometer and a doctor. One tells you the number. The other tells you what it means and what to do about it.

What this replaces (and what it doesn't)

Let me be direct: if you know your business (your offer, your margins, what a good customer looks like) this replaces most of what you're paying a PPC agency or consultant to do.

Reading search terms. Classifying intent. Drafting negatives. Flagging waste. Monitoring tracking health. Recommending structure changes. Writing RSA directions from real buyer language. Applying changes with audit trails.

That's 90% of the billable hours in a $3-5K/mo agency retainer. It's execution work billed at strategy rates.

This agent does all of it. Every morning. With memory.

Where it doesn't replace a human: knowing whether to run Google Ads at all. Knowing what your landing page should say. Knowing when a $300 CPA on comparison queries is a problem or an investment in a new market. That's strategy. That's the 10% that's genuinely hard.

But if you're paying someone $200/hr to scroll your search terms report? That's not strategy. That's maintenance. And you don't need a human for maintenance anymore.

The cost

Who this is for

• You're running Google Ads and your search terms report scares you (or you've stopped looking at it)

• You know something is wrong but every tool just shows you dashboards, not decisions

• You're an agency managing multiple accounts and need scalable search term analysis

• You've been burned by broad match expanding into garbage

Not for you if you want full auto-pilot or expect bid optimization. This is intent interpretation. A different and more fundamental problem.

Get the kit

The full system is open source:

15 OpenClaw skills. Read/Draft/Apply pipeline. The Query Interview framework. Intent Map that persists and compounds. Workspace memory across 12 persistent files.

Apply layer with dry run, audit trail, and instant undo. Connected mode via MCP or CSV export. Setup guide from install to first automated review.

No catch. No upsell. MIT license. Fork it.

If you'd rather have someone run this against your account and show you what your search terms are actually saying, that's what we do at Emerald. Reply and we'll talk.

Let’s go big,
Matt

P.S. That new client with the $4,200 in monthly waste? Cut in the first week. Their real buyer CPA: $112. Not the blended $179 the dashboard showed. No bid changes. No new keywords. Just stopped mixing unlike things together and let the numbers tell the truth.

P.P.S. If you missed the last three giveaways (the Meta Ads agent, the First 1000 Customers kit, and the SEO Kit) they're still free on GitHub.

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

Why does a video about '5 morning habits' get 2,000 views, but when the exact same creator posts about morning habits again, it gets 3.4 million views?"

Most people think the secret is finding a trending topic. But that's a lie. The topic is never the thing. The structure is the thing.

The biggest creators in your space aren't just lucky. They mechanically extract the structural DNA of viral videos.

They know every account worth knowing. They've watched the outlier videos (the ones that 10x'd someone's following overnight) and pulled apart what made them work.

They track which hook formats are spreading. They test at scale. They are, frankly, obsessed with their space in a way that compounds into an insurmountable advantage.

Most people can't sustain that. Or don't have time. So they guess.

On March 12th, Boring Marketer and I are building an Openclaw agent that does the obsessive part for you (automatically, every week). And we want you to come join us. You’re going to learn how to build a killer agent from scratch.

If you haven’t seen him blowing up in your feed, Boring Marketer is one of the biggest creators in the AI space.

He is Mr. Vibe Marketing. James (that’s his real name) has 100K followers on X, runs the Vibe Marketing community with 3,000 paying members, and is a regular on Greg Isenberg's pod. He's the person I trust to co-pilot this build.

We’re doing a live build. 90 minutes. You leave with a running system.

Your Openclaw agent will find outliers, analyze them, find out what works, and give you 10 concepts (in your voice).

I figured out how to extract that structure mechanically. The agent runs it every week on autopilot. I want your post analytics to look like this.

Sponsored

Why is everyone launching a newsletter?

Because it’s how creators turn attention into an owned audience, and an audience into a real, compounding business.

The smartest creators aren’t chasing followers. They’re building lists. And they’re building them on beehiiv, where growth, monetization, and ownership are built in from day one.

If you’re serious about turning what you know into something you own, there’s no better place to start. Find out why the fastest-growing newsletters choose beehiiv.

And for a limited time, take advantage of 30% off your first 3 months with code GROW30.

Start building for 30% off today.

What the pipeline actually produces

Twitter tweet

Every Monday, I get a Slack message that looks like this:

I didn't write those concepts. I didn't research those creators. I didn't sit down and try to figure out what made a video go viral.

The agent did all of it. I pick 3, we make them, and move on with my day.

That's what we're building.

How the pipeline works

Six stages. Each one feeds the next.

Stage 1: Virlo (the scanner)

Virlo watches creators in your niche and flags outlier performances.

Why outliers?

Last week it flagged a 51x outlier, a creator averaging 4,800 views, one video hit 249,000. Something about the structure of that content hit a nerve. That's what we want. Not their audience. Their blueprint.

We’re going to use the Virlo API. You can grab that here 👇

Stage 2: ScrapeCreators (the evidence pull)

Virlo spots the signal. But we need to get the content to analyze it.

ScrapeCreators fetches the full package: thumbnail, caption, stats, and creator context everything the AI needs to know why it worked.

I’m a big Scrapecreators fan and Adrian (guy who runs it) is great.

Stage 3: Content DNA (the breakdown)

This is where most people stop and start copying. "Oh, they did a video about morning routines. Let me do one about morning routines." That's the topic. The topic is not the thing.

Content DNA runs each video through Gemini 3 Flash via Openrouter and breaks down why the video worked across 7 dimensions:

• Topic — what the video is about at a high level

• Angle — the specific take that makes it unique

• Hook Structure — visual hook + text on screen + spoken opening line

• Story Structure — beat-by-beat narrative arc

• Visual Format — talking head / split-screen / green screen / etc.

• Key Visuals — the specific shots that carry the idea

• Audio — VO style, pacing, music, sound design

The output is what I call a "brick". That’s the structural blueprint stripped of the topic.

A brick can be applied to any niche, any voice, any product.

The brick from a fitness creator's viral video can power a SaaS founder's LinkedIn post. Because the structure that made people stop scrolling is universal. The topic is just the skin.

Stage 4: Ranking (the filter)

The agent compares all 10 breakdowns and scores each brick on portability and frequency.

Bricks showing up across 3+ creators are your highest-confidence bets. These aren’t format flukes, they're proven structural patterns.

That's what gets promoted to The Forge.

Stage 5: The Forge (the builder)

The Forge takes bricks and adapts them to your voice and niche. It uses three frameworks:

10 concepts come out the other end. Each one has the hook written, the structure mapped, the drive assigned, the traps applied. And these concepts are built on the structural DNA of content that already went viral, but adapted to sound like you.

Stage 6: Monday Drop (the cron)

A cron job fires the whole pipeline every Monday at 8am.

Virlo scans. Content DNA extracts. The Forge builds.

The results land in your Telegram (or Slack, or email. Whatever you want).

You pick 3. You create.

The loop compounds.

Week 1, the agent learns what's working in your space.

Week 4, it knows which Puppet Strings hit hardest for your audience.

Week 8, it's generating concepts that outperform anything you'd write from scratch. Because now it’s because it's built on 8 weeks of outlier data, not your gut feeling on a Monday morning.

Why we're building it live

I've given away four kits in four weeks. SEO. Meta Ads. First 1000 Customers. UGC Content Factory.

Twitter tweet

And every time, the same thing happens: my DMs fill up with "how do I actually set this up?" and "I cloned the repo but I'm stuck."

And I totally get it.

This tech is new. The piping is new. Openclaw workflows with cron scheduling and LLM-powered analysis… this isn't drag-and-drop, and we're not going to pretend it is.

The honest truth is that AI-native systems like this are still being figured out in real time.

The best way to learn them isn't to read a README. It's to watch someone build, follow the reasoning, see where things break, and understand why each piece connects.

What you leave with

Three things, all working before the session ends:

1. Your brand voice file. We run through the brand interview live. Eight questions. Your answers become the voice profile the agent uses for everything it writes. Takes 10 minutes. Changes everything about the output quality.

2. A running Virlo comet. Scanning your niche for outlier content. By the time the webinar ends, it's already collecting data. By next Monday, you have your first scan.

3. A Monday Drop cron, scheduled and firing. The full pipeline wired up. Virlo → Content DNA → The Forge → your Telegram. Set it and forget it.

You leave with a machine that's already running.

The live build

What: Build a Content Agent in 90 Minutes That Posts Like a 100K Creator

When: March 12, 2026

Who: Matt Berman + Boring Marketer

Format: Live build. You follow along and build yours at the same time.

What you need: A laptop, Openclaw already set up, and 90 minutes.
It’s helpful if you come in with a virlo key, a scrapecreators key, and an Openrouter key.

REGISTER FOR THE LIVE BUILD →

If you've ever cloned one of my kits and it's still sitting in a folder, this is how you actually get it running. After, there will be a repo that the Vibe Marketing crew will hand out.

If you want to join the Vibe Marketers community before the webinar (not required) you can do that here.

See you March 12th.

Matt

P.S. This is the first time I've done a live build instead of just dropping a repo. If it works (and it will), we're doing more of these. But this is the one where you get to be in the room while we figure out the format. First movers, etc.

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

I've been paying for Semrush for 15 years. Dashboards, reports, keyword tracking, CSV exports. Every client, every month. And I’m sick of it.

Then Covid hit and Google started crushing content that couldn't prove real expertise. Fake health articles, crypto scams… all wiped. Google called it YMYL. They wanted proof you knew what you were talking about.

Most agencies panicked. We started interviewing our clients. 15-30 minutes every week/month. What are customers asking? What do you know because you've actually done this work?

We'd take those answers, write articles with their real perspective in it, track rankings, build links, and repeat. It worked. Because Google wanted lived experience.

When AI content flooded the internet, same thing. Millions of generic articles. Google's AI Overview ate them alive. But the interview-based content? Kept ranking.

The problem was always scale. Keyword research, writing, link building, monitoring, site health. It all takes time.

So I built an OpenClaw agent that does everything except the interview. Four skills. Fourteen scripts. Runs every week on autopilot.

I'm giving the whole system away. Free.

┌─────────────────────────────────────────────────────────┐
│  ░█▀▀░█▀▀░█▀█░ ░█░░░█▀█░█▀█░█▀█                         │
│  ░▀▀█░█▀▀░█░█░ ░█░░░█░█░█░█░█▀▀                         │
│  ░▀▀▀░▀▀▀░▀▀▀░ ░▀▀▀░▀▀▀░▀▀▀░▀░░                         │
│                                                         │
│   DISCOVER ──→ WRITE ──→ LINK ──→ MONITOR ──→ COMPETE   │
│       ↑                                          │      │
│       └────────────── REPEAT ────────────────────┘      │
└─────────────────────────────────────────────────────────┘

Sponsored

What do these names have in common?

• Arnold Schwarzenegger

• Codie Sanchez

• Scott Galloway

• Colin & Samir

• Shaan Puri

• Jay Shetty

They all run their businesses on beehiiv. Newsletters, websites, digital products, and more. beehiiv is the only platform you need to take your content business to the next level.

🚨Limited time offer: Get 30% off your first 3 months on beehiiv. Just use code JOIN30 at checkout.

Start building for 30% off today.

Four Skills. One Loop.

Skill 1: SEO Agent (discovery + monitoring)

Connects to two data sources:

• Google Search Console for your actual ranking data. Not estimates. Your real positions, real clicks, real impressions.

• DataForSEO for keyword research, search volumes, competitor analysis, and SERP data. The API behind most SEO tools. Direct access costs a fraction of what Ahrefs charges.

Three scripts:

seo-discover pulls your GSC data and cross-references it with DataForSEO. Finds the strike zone: keywords where you're ranking 5-20. Close enough that one good supporting article could push you to page 1.

  GOOGLE RANKINGS
  ═══════════════════════════════════════

  Position 1-3    ██████████████  ← You're winning
  Position 4-5    ████████████    ← Defend these
  Position 5-20   ▓▓▓▓▓▓▓▓▓▓▓▓  ← STRIKE ZONE 🎯
                                    One article away
                                    from page one
  Position 20-50  ░░░░░░░░░░░░   ← Needs work
  Position 50+    ·············   ← Start over

seo-monitor snapshots your rankings weekly. What's climbing? What's dropping? It flags strike zone keywords ready to push and content losing ground before you notice.

seo-compete does competitor gap analysis. Give it a domain and it finds every keyword they rank for that you don't.

Skill 2: SEO Forge (content engine)

Most AI content tools write generic SEO garbage articles. They rank for 3 months, then drop. No voice. No experience. No reason to exist.

SEO Forge is different because of three things:

It interviews you first.

  ╔══════════════════════════════════════════╗
  ║   THE 8 QUESTIONS                        ║
  ║   ───────────────                        ║
  ║   1. What do you sell?                   ║
  ║   2. Who buys it?                        ║
  ║   3. What makes you different?           ║
  ║   4. What topics are you the expert on?  ║
  ║   5. What do customers keep asking?      ║
  ║   6. Share a sentence that sounds like   ║
  ║      "you"                               ║
  ║   7. What do competitors get wrong?      ║
  ║   8. What's your unfair advantage?       ║
  ╚══════════════════════════════════════════╝

From your answers, it builds a voice profile, audience file, and positioning doc. Everything it writes sounds like you, not ChatGPT.

It gets smarter every week. Quick check-in: What content performed well? Any new products? Questions customers keep asking? Content compounds because context compounds.

It builds topical authority. Every article fits into a hub-and-spoke structure. The agent maps your topic clusters, writes hub pages for broad keywords, and fills in spoke articles that strengthen the whole cluster.

                    ┌─────────────┐
            ┌──────→│  AI Marketing │←──────┐
            │       │  Automation   │       │
            │       │    (HUB)      │       │
            │       └──────┬───────┘       │
            │              │               │
     ┌──────┴───┐   ┌─────┴─────┐   ┌────┴──────┐
     │ AI Email  │   │ AI Social  │   │ AI Ad Copy │
     │ Marketing │   │ Scheduling │   │ Generation │
     │  (spoke)  │   │  (spoke)   │   │  (spoke)   │
     └──────────┘   └───────────┘   └───────────┘

Plus every article passes the "Only I Can Write This" test. If there's no sentence rooted in real experience, the agent flags it. That's the moat Google's AI Overview can't touch.

Skill 3: SEO Links (backlink engine)

Here's the truth nobody wants to say out loud: content without backlinks doesn't rank. You can write the best article on the internet and it will sit on page 3 forever if nobody links to it.

Most people know this and do nothing about it because link building is tedious. So I automated the discovery.

Five scripts:

link-mine pulls your competitor's backlink profile. Every site linking to them is a potential link for you. Resource pages, roundups, directories, guest posts. Prioritized by domain authority.

link-mentions finds pages that mention your brand but don't link to you. These are the easiest links to get. They already know who you are. A quick email converts at 15-25%.

link-broken crawls resource pages in your niche and checks for dead links. When you find a 404, you email the site owner: "Hey, this link on your page is broken. Here's my content that covers the same topic." You're doing them a favor.

link-internal audits your own site. Finds orphan pages with no internal links pointing to them. Suggests cross-links between related content. Run this every time you publish something new.

link-prospect finds "best tools," "top resources," and "ultimate guide" pages where you should be listed but aren't.

  ╔═══════════════════════════════════════════╗
  ║   WEEKLY LINK BUILDING                     ║
  ║   ────────────────────                     ║
  ║                                             ║
  ║   MON  link-internal                       ║
  ║        → fix orphan pages, add cross-links ║
  ║                                             ║
  ║   TUE  link-mine                           ║
  ║        → pull competitor backlinks         ║
  ║                                             ║
  ║   WED  link-mentions                       ║
  ║        → find unlinked brand mentions      ║
  ║                                             ║
  ║   THU  link-broken                         ║
  ║        → find broken links, pitch yours    ║
  ║                                             ║
  ║   FRI  link-prospect                       ║
  ║        → find resource pages to get on     ║
  ╚═══════════════════════════════════════════╝

The agent handles all the discovery. You (or your team) handle the outreach. 30 minutes a day. (I’m going to automate this part later too so stay subscribed)

Skill 4: SEO Health (technical monitoring)

Rankings depend on a solid foundation. A page with perfect content and strong backlinks will still rank poorly if it takes 6 seconds to load.

Three scripts:

health-speed hits Google's PageSpeed Insights API (free) and checks Core Web Vitals. LCP, INP, CLS, FCP, TTFB. Tracks scores over time and flags regressions. "Your LCP got 400ms worse this week". You'd want to know that before your rankings drop.

health-crawl audits your site for broken internal links, redirect chains, missing meta tags, duplicate titles, and mixed content. Finds the stuff that silently tanks your SEO.

health-images checks every image on your top pages. Missing alt text, oversized files, wrong formats, missing dimensions (the #1 cause of layout shift). Tells you exactly what to fix.

Run all three every Monday morning. The agent reads the output, prioritizes fixes, and tells you what matters.

What a Week Looks Like

  ╔═══════════════════════════════════════════════╗
  ║   YOUR WEEK ON AUTOPILOT                       ║
  ║   ─────────────────────                        ║
  ║                                                 ║
  ║   MON  discover + monitor + health checks      ║
  ║        → "12 opportunities. 3 climbers.        ║
  ║           LCP is good. 2 broken links found."  ║
  ║                                                 ║
  ║   TUE  forge writes 2 articles                 ║
  ║        → drafts in your inbox to review        ║
  ║                                                 ║
  ║   WED  internal links + unlinked mentions      ║
  ║        → "Added 8 cross-links. Found 4         ║
  ║           sites mentioning you without links."  ║
  ║                                                 ║
  ║   THU  competitor backlinks + broken links     ║
  ║        → "15 backlink targets. 3 broken link   ║
  ║           opportunities ready to pitch."        ║
  ║                                                 ║
  ║   FRI  compete + resource prospecting          ║
  ║        → "8 keywords they have, you don't.     ║
  ║           6 resource pages to submit to."       ║
  ║                                                 ║
  ║   YOU  review + publish + outreach. 30 min/day ║
  ╚═══════════════════════════════════════════════╝

The loop compounds.

Week 1, the agent discovers opportunities and writes content.

Week 2, it monitors what indexed and builds links to the new content.

Week 3, it runs competitor gaps and finds new angles.

By week 4, your first articles are moving up and the agent is already three steps ahead.

The Loop (This Is the Real Magic)

Each skill is useful alone. Together, they create something that compounds.

Week 1: Discover 20 keyword opportunities
         Write 3 articles targeting the best ones
         Add internal links between new + existing content
         Run site health check, fix 2 broken links

Week 2: Monitor rankings — 2 articles indexed, 1 in strike zone
         Write supporting content for the climber
         Mine competitor backlinks — 15 outreach targets
         Send 5 unlinked mention emails — 2 link back

Week 3: Check GSC — first article moved from position 15 to 9
         Run competitor gap, find 8 new opportunities
         Broken link outreach — 1 resource page adds your link
         Submit to 3 "best tools" pages

Week 4: Write 3 more articles from gap analysis
         First article hits position 6
         Organic traffic up 34%
         Backlink profile growing every week

Repeat forever.

Nobody does this manually. Not consistently. Not across content, links, AND technical health simultaneously. The agent never gets distracted. It runs every week whether you're on vacation or buried in client work.

And because it monitors what's actually working (not what you think should work), it gets better over time. Week 4 is smarter than week 1. Week 12 is running a machine.

The Cost

  ┌────────────────────┐     ┌──────────────────────┐
  │  THE OLD WAY        │    │  THE AGENT WAY       │
  │  ──────────         │    │  ──────────          │
  │  Semrush: $200/mo   │    │  DataForSEO: $50/mo  │
  │  Ahrefs:  $200/mo   │    │  GSC: free           │
  │  Surfer:  $89/mo    │    │  PageSpeed API: free │
  │  Jasper:  $80/mo    │    │  OpenClaw: free      │
  │  Writer:  $5k/mo    │    │  SEO Kit: free       │
  │                     │    │                      │
  │  Total: $5,569/mo   │    │  Total: $50/mo       │
  │  Clients: 4         │    │  Clients: 12         │
  └────────────────────┘     └──────────────────────┘

Skip DataForSEO entirely and use the web search fallback. Less data, still works (technically but I’d go for the data).

Total: $0.

"What If I Don't Know SEO?"

That's the point.

The agent knows SEO. It knows what a strike zone keyword is. It knows how to read SERP competition. It knows that supporting content with internal links pushes the parent piece higher. It knows to check search intent before writing.

You know your business. You know what customers ask. You know what makes your product different. You know the stories only you can tell.

The interview connects those two things. You bring the expertise. The agent brings the SEO.

What You're Getting

Four skills. Fourteen scripts. The full stack.

• seo-agent/ — discover, monitor, compete (keyword intelligence)

• seo-forge/ — brand interview, content engine, weekly refinement

• seo-links/ — competitor mining, unlinked mentions, broken links, internal audit, resource prospecting

• seo-health/ — PageSpeed, crawl audit, image optimization

• seo-checklist/ — meta tags, schema markup, llms.txt, topical authority reference

• SOUL.md, AGENTS.md, setup guide, everything

Clone it. Run seo-check.sh to verify your connections. Let the agent interview you. Run your first discovery. Check your site health.

Set it on a schedule and forget about it. Check back in a month.

If you build something cool with it, reply to this email. I read every one.

See you next Tuesday. Hit me up or hit reply.

Matt

P.S. If you missed the last two Openclaw agents (the Meta Ads agent and the First 1000 Customers kit), they're still free on GitHub.

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

I Woke Up Monday To This Openclaw Message

I didn't touch Ads Manager. I didn't open my laptop. I read the text as I got my kids ready for school and typed "approved."

That's an OpenClaw agent doing the job. Running my ads account.

I'm giving the whole thing away free. The agent config, three skill, the setup guide. All of it. If you've been waiting for someone to hand you a working AI ad manager, this is it.

Let me show you how it works.

┌───────────────────────────────────────────────┐
│ ░█▀█░█▀█░█░█░█▀▀░█▀▄░ ░█▀▀░█░█░▀█▀░█▀▀░▀█▀    │
│ ░█▀▀░█░█░█▄█░█▀▀░█▀▄░ ░▀▀█░█▀█░░█░░█▀▀░░█░    │
│ ░▀░░░▀▀▀░▀░▀░▀▀▀░▀░▀░ ░▀▀▀░▀░▀░▀▀▀░▀░░░░▀░    │
│          FROM ASKING → TO ACTING              │
└───────────────────────────────────────────────┘

Investor-ready updates, by voice

High-stakes communications need precision. Wispr Flow turns speech into polished, publishable writing you can paste into investor updates, earnings notes, board recaps, and executive summaries. Speak constraints, numbers, and context and Flow will remove filler, fix punctuation, format lists, and preserve tone so your messages are clear and confident. Use saved templates for recurring financial formats and create consistent reports with less editing. Works across Mac, Windows, and iPhone. Try Wispr Flow for finance.

Try Wispr Flow

Last time was questions. This is action.

A few months ago I wrote about cutting Meta ads reporting to 45 seconds using an MCP server and plain English questions. That piece got a ton of response. People loved the idea of asking their ad account questions instead of clicking through the Ads Manager nightmare.

But asking questions isn't the gig (just a part).

The job is: pause the loser at 3am before it burns $200. Scale the winner before the window closes. Upload 50 creatives perfectly named, perfectly organized, without a single mistake. Shift budget from dying ad sets to climbing ones.

That's what this agent does.

We didn't ditch MCP. My team is still on it. We deployed it to Railway, added publishing capabilities, got the whole workflow running. MCP is solid. But here's the thing: every MCP call eats your context window. When you want a recurring schedule that runs at 7am every morning (across dozens of ad accounts) whether you're awake or not, that token burn adds up fast. It also requires an active session.

OpenClaw's heartbeats and cron solve that. The agent runs on schedule. No active session needed. Barely touches context. MCP is still in the stack for the interactive stuff. This is the evolution, not a replacement.

  ╔══════════════════════════════════════╗
  ║   THE 5 DAILY QUESTIONS              ║
  ║   ─────────────────────              ║
  ║   1. Am I on track?                  ║
  ║   2. What's running?                 ║
  ║   3. How's performance?              ║
  ║   4. Who's winning / losing?         ║
  ║   5. Any fatigue?                    ║
  ╚══════════════════════════════════════╝

What the agent actually does

Here's what's running in my account right now:

Every morning at 7am, my OpenClaw agent fires the meta-ads skill. That skill wraps social-cli, which is an open-source CLI that talks directly to Meta's Marketing API. No middleware. No protocol overhead. Just commands.

The skill runs 5 questions against live data. Then it acts.

Question 1: Am I on track?

Checks spend pacing against daily budget. If you're 60% through the day with 20% of budget spent, something's wrong. If you're blowing through budget by noon, something else is wrong. The agent flags both and doesn't wait for you to notice.

Question 2: What's running?

Pulls all active campaigns, status, current spend. Gives you a clean picture of what's actually live versus what you thought was live. You'd be surprised how often those two things don't match.

Question 3: How's performance?

7-day metrics by campaign. ROAS, CPA, CPL, CTR. Sorted by what matters. Not buried in Ads Manager columns you have to rearrange every time.

Question 4: Who's winning? Who's losing?

Ad-level data, ranked. Best to worst. The agent shows you the stack. The winners are obvious. So are the ones bleeding money. Then it acts on the obvious ones without waiting to be told.

Question 5: Any fatigue?

This is the one most people miss. CTR trends over time. Frequency (how many times the same person sees the same ad). CPC climbing without obvious cause. These are the early signals of creative fatigue. By the time your CPA spikes, you've already wasted a week. Catching frequency creeping past 3 before the CPA moves saves real money.

What a morning looks like

7:02am. Telegram message from the agent:

I read that in 90 seconds. Reply "approved" on the scaling rec. Ask about the landing page. Done.

Total time managing ads that day: under 5 minutes.

The architecture

Here's how the pieces connect.

OpenClaw agent. This is the brain. It runs on schedule via cron and heartbeats. It interprets data, applies decision rules, sends me the brief. It's what turns a pile of ad metrics into "here's what happened and here's what I did about it."

OpenClaw is open source. That's important. You own the whole thing. Not renting SaaS. Not hoping some platform doesn't change their pricing. Your agent, your rules.

Three skills. This is the knowledge layer.

• The meta-ads skill runs the 5 daily questions and handles pause/resume/budget actions.

• The ad-creative-monitor tracks creative health over time, catching fatigue days before your CPA spikes.

• The budget-optimizer ranks campaign efficiency and tells you where to shift spend.

Together they're the difference between raw data and an actual decision. They're what make the agent behave like someone who actually knows paid media.

Clone the repo. The agent picks up all three automatically.

social-cli. This is the connection to Meta. Open source CLI for the Graph API. Handles token refresh, pagination, rate limits. All the stuff that makes Meta's API miserable to work with directly. Honestly go follow and give some love to Vishal Ojha who open sourced this. I was working on something similar, but I saw he open sourced this, I dug under the hood and it’s fantastic (better than what I had).

social marketing insights --preset last_7d --level ad
social marketing pause ad AD_ID
social marketing set-budget adset ADSET_ID --daily-budget 5000
social marketing status

Those four commands are most of what you need. The agent knows how to use them. The skill tells it when.

Cron and heartbeats. This is what makes it actually run. A cron job fires at 7am. The OpenClaw heartbeat system handles the scheduling without requiring an active session. This is why we're not doing it through MCP alone. The agent wakes up, does its job, sends the report, goes back to sleep. Your context window stays clean.

The full picture:

OpenClaw agent
    → meta-ads skill (daily checks, actions)
    → ad-creative-monitor (fatigue tracking)
    → budget-optimizer (spend efficiency)
        → social-cli
            → Meta Marketing API

Cron triggers the chain. The brief lands in your channel of choice.

The action commands

The skill doesn't just report on data. It does stuff for you!

Pause bleeders. Any ad running more than 3x target CPA for 48+ hours gets auto-paused. No approval needed. The agent notifies you after. You can always resume, but the bleeding stops.

Resume winners. If an ad that was previously paused has a changed context (new audience, refreshed creative) the agent can resume it on instruction.

Shift budget. This one's approval-required for now. The agent recommends a budget shift with the data behind it. You say yes or no in two seconds. It executes.

Fatigue checks. When frequency or declining CTR triggers the threshold, the agent flags the creative and writes the observation to a workspace file. Over time you build a log of what fatigued when and why. Pattern recognition you can actually use.

Custom reports. Ask it anything. "What's my CPL trend over the last 30 days?" "Which creatives are still performing after day 14?" It pulls the data and builds the answer.

  ┌─────────────────────────────────────┐
  │  DATA SIDE        CREATIVE SIDE     │
  │  ──────────       ─────────────     │
  │  social-cli  ───▶  StealAds         │
  │  (what's          (make more        │
  │   winning)         of it)           │
  │                                     │
  │        THE FULL LOOP  ↺             │
  └─────────────────────────────────────┘

The creative side (and how StealAds happened)

The agent handles data and decisions. But ads still need creative. And creative at scale was the bottleneck I kept hitting.

Quick backstory if you missed it. A few weeks ago I posted an n8n automation that pulled competitor ads, analyzed the creative patterns, and generated new variations. It went viral. 4,000 DMs later it was clear I wasn't the only one with this problem.

Twitter tweet

So I built it into a real platform. StealAds. It's live. I'm accepting new batches of early access users every week.

Here's what it does: you feed it competitor ads or winning creative from your own account. It analyzes the visual DNA. The layout, the color psychology, the hook style, the format. Then it generates new variations adapted to your brand. Not templates. Not Canva. AI-generated creative that actually looks like something a paid media team would produce.

Between social-cli pulling performance data and flagging what's winning, and StealAds generating creative based on those signals, you start to see the loop. The agent tells you what's working. StealAds helps you make more of it. Fast. We're building the OpenClaw skills and API integrations for StealAds now so the loop gets even tighter: agent sees a winner, drafts a creative brief, StealAds generates the variants. Not quite there yet, but that's where this is heading.

If you want early access: stealads.ai or REPLY to this email. I’m only opening up a handful of slots every week during early access.

  ⚠️  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
      META'S AI OPTIMIZES FOR META.
      YOUR AI OPTIMIZES FOR YOU.
      CHOOSE ACCORDINGLY.
  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ⚠️

The Manus problem

Meta just acquired Manus AI. And now they're rolling out AI inside Ads Manager. My team isn't fans (just yet).

Here's the thing about Meta building AI into your ad management: Meta does not optimize for you. They optimize for Meta.

Twitter tweet

They want you spending more. They want broader audiences because broader audiences mean more inventory sold. They want you running more placements because more placements means more revenue for them. Their AI will always point in that direction.

And look at their track record as a platform steward.

• They killed organic reach.

• They changed attribution windows without warning.

• They sunset tools entire marketing stacks were built on.

• They changed their ad policies in ways that cratered campaigns overnight.

Meta has not proven to be a trustworthy cultural or brand steward. That's not even a hot take. That's history.

So when Meta says "let our AI manage your ads for you," the smart question is: manage them toward what outcome? Yours or theirs?

Your AI agent optimizes for your target CPA. Your cash flow constraints. Your brand guardrails. It has no incentive to burn your budget faster.

The winning move isn't choosing between Meta's AI and your own. Use both. Let Meta handle targeting and delivery (they're genuinely good at that now). Use your own agent to manage the strategy layer on top.

You keep the judgment. They run the auction.

  ┌─────────────────────────────────────┐
  │  GET RUNNING IN 30 MIN              │
  │                                     │
  │  Step 1: Install social-cli   5min  │
  │  Step 2: Clone the agent kit   2min  │
  │  Step 3: openclaw start        1min  │
  │  Step 4: Set up morning cron  5min  │
  │  Step 5: Read your first brief ☕   │
  └─────────────────────────────────────┘

Getting started

Step 1: Install social-cli (15 minutes)

npm install -g @vishalgojha/social-cli
social auth login
social marketing accounts
social marketing set-default-account act_YOUR_ID

Test it:

social marketing status
social marketing insights --preset last_7d --level campaign

First time you see your ad data in a terminal is a thing. It hits different.

Step 2: Get the OpenClaw agent and skill (free)

I'm giving away the complete setup. The agent config, the meta-ads skill with all 5 daily questions and action commands, and the setup guide. Everything you need to run this in your own account.

grab it here: https://github.com/TheMattBerman/meta-ads-kit

Step 3: Drop the skill into your workspace

The skill goes in your OpenClaw workspace. Point it at your ad account. You're done. The agent picks it up automatically.

Step 4: Set up the morning cron

One cron job. Fires at 7am. Runs the 5 questions. Sends the brief to wherever you want it. Telegram, Slack, email. Your call.

Step 5: Start in review mode

For the first week, the agent recommends. You approve. Get comfortable with what it's seeing and how it's reasoning. Then graduate to auto-pause for the obvious bleeders. Keep the budget shifts in approval mode as long as you want.

You don't have to hand over full autonomy on day one. Start with visibility. Earn trust in both directions.

The honest version of what this replaces

Not a media buying team. Not creative strategy. Not positioning. Not the calls that actually require judgment.

What it replaces:

The 20 minute sessions clicking through Ads Manager hoping you don't miss something. The end-of-week scramble to figure out why your CPL jumped and whether it was day-of-week or creative fatigue or audience overlap. The extra $1200 that burned overnight because you didn't catch the CPA spike before bed.

That's not strategy brother. That's maintenance. And maintenance is a terrible use of your time.

The 10% that's actually hard, knowing when to go broad versus when to tighten targeting, knowing which creative angle to test next, knowing when the data is noise versus signal, that's what you should be spending your brain on.

Not clicking.

Get the agent (free)

I'm giving away the whole stack. A complete OpenClaw agent with three skills, decision logic, and everything configured. Clone it, authenticate, run it.

You get:

• Complete OpenClaw agent ready to run (SOUL.md, AGENTS.md, clawdbot.json, the works)

• meta-ads skill with all 5 daily questions and action commands

• ad-creative-monitor skill that tracks creative health over time and catches fatigue before your CPA spikes

• budget-optimizer skill that ranks campaign efficiency and recommends where to shift spend

• Action commands: pause bleeders, resume winners, shift budget

• Morning brief template that lands in Telegram, Slack, or wherever you want it

• Setup guide from install to first automated report

• social-cli install instructions and auth setup

Takes about 30 minutes to get running. Then it runs.

If you'd rather have someone build this into your account and show you what your data is actually saying, that's what we do at my agency. Reply and we'll talk.

Let's build,

Matt

P.S. The Ads Manager UI is still one of the worst pieces of software I've ever regularly used. This doesn't make it better. It makes you not have to use it.

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

I Have an AI Agent That Finds Me Customers

Let me guess.

You spent more time on "prospecting" this week than building your actual product. LinkedIn. Spreadsheets. Cold emails. Follow-ups you forgot to send.

Your pipeline is a mess. Your product is good. And somewhere in the back of your head, you know you can't outwork a funded team's full-time outbound machine.

Good news: you don't have to.

I built an AI agent that finds me customers. Text it from my phone. It scrapes warm leads, enriches emails, writes personalized outreach, loads sequences. Runs every 4 hours.

Built it last week. 3 discovery calls booked this week. High-ticket.

And I'm giving the whole system away. Free. Open source. The full stack.

Because here's the thing: the founders who figure out AI-native GTM first are going to eat everyone else's lunch. And I'd rather that be you than another VC-backed team with unlimited runway.

Here's exactly how it works.

When it all clicks.

Why does business news feel like it’s written for people who already get it?

Morning Brew changes that.

It’s a free newsletter that breaks down what’s going on in business, finance, and tech — clearly, quickly, and with enough personality to keep things interesting. The result? You don’t just skim headlines. You actually understand what’s going on.

Try it yourself and join over 4 million professionals reading daily.

Check it out

The GTM Problem Is Getting Worse

Here's something nobody's talking about: GTM engineers are becoming the most valuable hires in tech.

Twitter tweet

I'm serious. The best ones are getting gobbled up by funded companies paying $180K, $200K, $250K. And that number is going UP, not down.

Why? Because it's the hardest skill in the game right now.

A great GTM engineer is:

• Technical enough to build systems

• Marketing-savvy enough to write copy that converts

• AI-native enough to leverage the new tools

• Strategic enough to prioritize ruthlessly

That's a rare combination. The people who have it know their worth. The companies that need it are bidding against each other.

Twitter tweet

And if you're a bootstrapped founder? You're priced out.

So what do you do?

You build something 90% as effective. For 1% of the cost.

"Cool, But What Do I Actually Build With It?"

I hear this question constantly.

OpenClaw gives you an AI agent you can message from anywhere. Phone, desktop, wherever. It runs code, calls APIs, browses the web, coordinates tasks. Like having a technical co-founder available 24/7.

But "you can do anything" isn't helpful. You need a specific, valuable thing.

Twitter tweet

So I built the most valuable thing I could think of for a bootstrapped founder: a complete GTM pipeline that turns engagement mining into booked meetings.

A full production system that booked me 3 calls in its first week.

What Having a GTM Agent Actually Feels Like

This is hard to explain until you experience it.

I used to spend 3 hours a day on prospecting. Browsing LinkedIn (and I hate LinkedIn). Copying names into spreadsheets. Guessing email addresses. Writing the same mediocre cold email 50 times. Forgetting to follow up.

Now I text my phone:

And walk away.

My OpenClaw agent:

• Scrapes 20 relevant posts

• Extracts everyone who engaged (156 people)

• Finds their verified emails (83% hit rate)

• Scores them against my ICP (12 A-tier, 34 B-tier)

• Writes personalized outreach based on their comments

• Loads everything into Instantly

• Starts the sequence tomorrow morning

I touched nothing after the initial message.

When someone replies and books a call? Another agent skill fires. Perplexity does a deep research dive. I get a 1-page briefing 30 minutes before the meeting: their background, recent posts, company situation, talking points, objections to expect.

This is what enterprise sales teams have. Now I have it in my pocket.

The 8 Skills That Replace a $200K Hire

This system chains 8 OpenClaw skills together:

┌─────────────────────────────────────────────────────────────────┐
│                     FIRST 1000 KIT PIPELINE                     │
└─────────────────────────────────────────────────────────────────┘

  ┌──────────┐    ┌──────────┐    ┌──────────┐    ┌──────────┐
  │  MINE    │───▶│  ENRICH  │───▶│  VERIFY  │───▶│  SCORE   │
  │ LinkedIn │    │  Emails  │    │  Clean   │    │  0-100   │
  └──────────┘    └──────────┘    └──────────┘    └──────────┘
       │                                               │
       │  ← Warm leads from                           │
       │    post engagement                           ▼
                                              ┌──────────┐
  ┌──────────┐    ┌──────────┐                │  WRITE   │
  │  CALLS   │◀───│   LOAD   │◀───────────────│  Emails  │
  │ Booked!  │    │ Instantly│                └──────────┘
  └──────────┘    └──────────┘

  Your phone ──▶ "Find me customers" ──▶ Meetings on calendar

You message your agent. It coordinates everything. You get meetings.

The secret weapon: Signal Scanner.

Before outreach, the system runs Perplexity deep research on every A-tier lead. It checks:

• Did they just raise funding? (They have budget)

• Are they hiring growth/marketing roles? (They're investing)

• Any recent press or product launches? (They're active)

A lead that fits your ICP AND just raised a Series A? That's not just A-tier. That's Priority 1. Reach out today.

Why This Works Better Than "AI Prospecting Tools"

Cold outreach is dying. Response rates are cratering…

But there's a loophole: people who engage with niche content are raising their hand.

When someone comments on a post about rising CPMs, they're telling you "I care about this topic." But nobody's reaching out to them in context.

The LinkedIn Mining skill captures that signal:

1. Monitors posts on topics you care about

2. Scrapes everyone who engages

3. Captures the context ie what they said, what post they responded to

4. Prioritizes commenters (higher intent than passive likers)

Now your outreach isn't as cold. It's:

That's a completely different conversation than "Hi {FirstName}, I noticed you're a VP of Marketing..."

Warm context. At scale. From your phone.

The Stack

Here's what's running under the hood:

┌────────────────────────────────────────────────────────────────┐
│                        THE STACK                               │
├────────────────────────────────────────────────────────────────┤
│  OpenClaw          │  Your AI agent (orchestration)  │  Free  │
│  RapidAPI          │  LinkedIn scraping + profiles   │  ~$30  │
│  Hunter.io         │  Email finding + verification   │  ~$50  │
│  Instantly.ai      │  Email sequences                │  $30   │
│  Perplexity        │  Deep research before calls     │  $20   │
├────────────────────────────────────────────────────────────────┤
│  TOTAL                                               │  ~$130 │
└────────────────────────────────────────────────────────────────┘

For context, that's one client dinner. Except this books the meetings that lead to client dinners.

The whole system runs through OpenClaw. Message it from your phone, computer, wherever. One command kicks off everything:

"Mine Meta ad costs DTC, 20 posts, score and load to February campaign"

[10:31:02] Mining 20 posts...
[10:31:45] Found 156 engagers
[10:31:46] Scored: 12 A-tier, 34 B-tier, 110 skipped
[10:31:52] Generated 46 email sequences
[10:31:58] Loaded to Instantly campaign "February"

Done. 46 leads queued. First emails send tomorrow 9am.

The Math

┌────────────────────────────────────────────────────────────────┐
│                  HIRE A GTM    vs    AI AGENT                  │
├────────────────────────────┬───────────────────────────────────┤
│  $200,000/year (and rising)│  $130/month ($1,560/year)         │
│  3-6 months to ramp        │  1 afternoon to set up            │
│  60% failure rate          │  Cancel anytime                   │
│  Competing for scarce hire │  Build it yourself                │
│  Managing another human    │  Runs while you sleep             │
│  PTO, benefits, meetings   │  Messages from your phone         │
├────────────────────────────┼───────────────────────────────────┤
│       ❌ BIDDING WAR        │        ✅ AGENT                    │
└────────────────────────────┴───────────────────────────────────┘

128x cost reduction.

But here's what really matters: you're not waiting 6 months to see if your hire works out. You see pipeline in week one. I saw it in week one. 3 discovery calls from a system I built last Tuesday.

What the Output Actually Looks Like

Real flow from last week:

Input: "Find DTC founders complaining about ad spend and ROAS"

45 minutes later:

┌─────────────────────────────────────────────────────────────┐
│                    PIPELINE OUTPUT                          │
├─────────────────────────────────────────────────────────────┤
│  Leads extracted:     47                                    │
│  Verified emails:     39  (83% hit rate)                    │
│  A-tier (80+):        12  ← Priority outreach               │
│  B-tier (60-79):      18  ← Sequence outreach               │
│  C-tier (40-59):       9  ← Lower priority                  │
│  Skipped:              8  ← Not ICP fit                     │
├─────────────────────────────────────────────────────────────┤
│  Sequences generated: 30                                    │
│  Loaded to Instantly: ✅                                    │
│  First send:          Next morning 9am                      │
└─────────────────────────────────────────────────────────────┘

Result: 3 discovery calls booked by end of week.

Who This Is For (And Who It's Not)

This is for you if:

• You're a bootstrapped founder who needs pipeline yesterday

• You can't afford (or can't find) a $200K GTM engineer

• You're a solo founder wearing every hat including sales

• You're an agency owner doing outbound for yourself or clients

This is NOT for you if:

• You have unlimited budget and want to throw bodies at the problem

• You're not willing to spend 10 minutes setting up API keys

• You expect 100% automation with zero oversight (you still setup, steer, review and approve)

This is a force multiplier, not a replacement for judgment. You're still the strategist. The system just removes the crap work between "who should I target?" and "meetings on the calendar."

Get the Kit

I built this for myself. Then I realized it's exactly what every bootstrapped founder needs (and exactly what people keep asking about OpenClaw).

The full system is open source:

👉 github.com/TheMattBerman/first-1000-kit

• All 7 OpenClaw skills (full source)

• Setup documentation (10 minutes to configure)

• ICP scoring templates

• Email sequence frameworks

• The exact prompts that make it work

No catch. No upsell. MIT license. Fork it. Build on it.

This is what you do with OpenClaw. This is one answer to "cool, but what do I actually build?"

You build the thing that gets you customers.

GTM Engineers Are Amazing. And Unaffordable.

Let me be clear: I'm not saying GTM engineers are obsolete.

Twitter tweet

The opposite. They're becoming the most valuable position in tech. Technical + marketing + AI-native is the killer combo, and the people who have it are commanding higher salaries every quarter.

If you can afford one, hire one. They're worth it.

But most bootstrapped founders can't compete in that bidding war. So you do what bootstrapped founders have always done: you build it yourself.

This kit is 90% of a GTM engineer, for about 1% of the cost.

For your first 1000 customers? That's enough.

Go build. Go big.

— Matt Berman

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

You're paying $300-500 per UGC video.

Maybe more. The good creators charge $800+.

And here's what you get for that money:

• 1 video

• 1 script interpretation

• 1 take (maybe 2 if you're lucky)

• 1 creator who may or may not match your audience

• 3-5 day turnaround

Want variations? Pay again. Different demographic? Pay again. A/B test hooks? You guessed it.

The economics are broken.

Last week, Kling 3.0 dropped. And buried in the announcement was a feature that changes everything.

Twitter tweet

It's called "Elements."

Upload a reference image of a person. The AI locks their identity across every video you generate. But here's the part most people miss: Elements isn't just for faces. Upload your SaaS dashboard, your app screen, your product packaging (and it appears in the video as a locked element too). Your AI creator actually using your actual product.

Combine that with native lip-sync in 5 languages, multi-shot storyboards (up to 6 shots), and 4K output...

You're not generating videos anymore. You're building a creator roster with your real product baked in.

I’m giving away a Skill you can use for free to make your UGC army. Here's exactly how I'm using it.

Ship the message as fast as you think

Founders spend too much time drafting the same kinds of messages. Wispr Flow turns spoken thinking into final-draft writing so you can record investor updates, product briefs, and run-of-the-mill status notes by voice. Use saved snippets for recurring intros, insert calendar links by voice, and keep comms consistent across the team. It preserves your tone, fixes punctuation, and formats lists so you send confident messages fast. Works on Mac, Windows, and iPhone. Try Wispr Flow for founders.

Start flowing free

THE UNLOCK: Character & Product Persistence

Before Elements, AI video had a fatal flaw for UGC: consistency.

You'd generate a testimonial video. Great. Generate another with the same "person"? Completely different face. Different vibe. Different everything.

No brand would use that. UGC works because people recognize the creator. Trust compounds across videos.

Elements solves this. And it goes further than most people realize.

Twitter tweet

Upload one reference image of a person. The model creates an identity lock. Now that exact person (same face, same features, same vibe) appears in every video you generate.

But Elements supports multiple locked references per video. Upload your creator and your product (a SaaS dashboard screenshot, an app screen, product packaging) and both appear consistently. Your AI creator scrolling through your actual app. Holding your actual product. Reacting to your actual interface.

Same product. Different scripts. Different settings.

That's the foundation of a UGC factory.

THE SYSTEM: Director-First Pipeline

Here's the architecture I've been testing. The key insight: visuals before script. Design how each shot looks and moves first, then write dialogue that fits. The script serves the visuals, not the other way around.

UGC CONTENT FACTORY — DIRECTOR-FIRST PIPELINE

┌─────────────────────────────────────────┐
│  1. CHARACTER DESIGN                    │
│  → Generate reference images (Nano      │
│    Banana) matching your target demo    │
│  → Lock character in Kling Elements     │
└─────────────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────┐
│  2. VISUAL STORYBOARD                   │
│  → Design shot-by-shot visual sequence  │
│  → Variable durations (3s/4s/5s)        │
│  → Kling prompt engineering             │
│  → NO dialogue yet — visuals first      │
└─────────────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────┐
│  3. SYNCHRONIZED SCRIPT                 │
│  → Write dialogue fitted to shot        │
│    durations (~2.5 words/sec)           │
│  → Natural cadence + filler words       │
│  → Script serves the visuals            │
└─────────────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────┐
│  4. GENERATE + MULTIPLY                 │
│  → fal.ai Kling 3.0 with character lock │
│  → Multi-shot storyboard                │
│  → Native lip-sync                      │
│  → Variant multiplication               │
└─────────────────────────────────────────┘

Let me break down each step.

STEP 1: Build Your Creator Roster

First, you need characters that match your target audience.

(As always) I’m using Nano Banana Pro (Google's image model via Replicate) to generate reference images.

The secret: Most AI image prompts create obviously fake people. The fix is adding imperfections deliberately.

The Photorealistic Prompt Formula:

Raw iPhone [model] photo, [shot type] of [character details],
[hair with imperfections], [natural expression], [casual clothing],
[activity/pose], [specific location], [natural lighting + time],
candid moment, unfiltered, authentic Instagram aesthetic, f/1.8,
shallow depth of field, slight grain

Physical details: visible skin texture and pores, natural under-eye
area, flyaway hairs, subtle facial asymmetry, realistic hands
(5 fingers, natural pose)

Negative prompt: CGI, 3D render, perfect skin, plastic, beauty
filter, symmetrical, studio lighting, fake, artificial, model pose

Example for a DTC skincare brand:

Raw iPhone 15 Pro photo, front-facing selfie of 32 year old woman,
shoulder-length brown hair slightly messy, warm genuine smile with
slight laugh lines, wearing casual cream sweater, sitting at kitchen
counter, morning window light creating soft shadows, candid moment,
unfiltered, authentic Instagram aesthetic, f/1.8, slight grain

Physical details: visible skin texture and pores, natural under-eye
area, flyaway hairs, subtle facial asymmetry, realistic hands

Negative prompt: CGI, 3D render, perfect skin, plastic, beauty filter,
symmetrical, studio lighting, fake, artificial, influencer pose

Generate 5 variations. Pick the best. That's your creator.

The key: Natural, not perfect. Add imperfections deliberately. Flyaway hairs, under-eye texture, asymmetry. The negative prompt does half the work by telling the AI what not to do.

Once you have your reference, you pass it to the Elements system via fal.ai's API. Now every video you generate uses that locked identity.

I built a roster of 6 characters:

• Female, 28-35, casual lifestyle (general consumer)

• Female, 22-28, trendy/energetic (Gen-Z products)

• Male, 30-40, professional casual (B2B/productivity)

• Male, 25-35, fitness aesthetic (health/wellness)

• Female, 35-45, put-together mom (family/home products)

• Male, 22-30, creator aesthetic (tech/apps)

Different products call different creators from the roster.

STEP 2: Design Your Visual Storyboard (Director-First)

Key principle: visuals before script. Design how each shot looks and moves first, then write dialogue to fit. Most people do this backward—they write a script and try to make visuals match. That leads to talking-head videos where nothing changes visually.

This is where Kling 3.0's multi-shot feature shines. You can define up to 6 distinct shots in one generation. The AI handles transitions, shot composition, and character continuity automatically.

Variable shot durations: Each shot can be 3s, 4s, or 5s. Hooks should be punchy (3s). Demos need room to breathe (5s). Total must sum to your target duration.

Here's my testimonial storyboard template:

SHOT 1 (3s): HOOK
- Character at kitchen counter, picks up phone
- Selfie angle, looks at camera
- Expression: genuine excitement
- [Dialogue TBD — visuals first]

SHOT 2 (5s): SETUP + PRODUCT
- Same setting, character holds up product
- Shows product clearly, closer framing
- Expression: engaged, sharing
- [Dialogue TBD]

SHOT 3 (4s): PAYOFF
- **Stands up**, demonstrates product
- Camera angle change
- Expression: genuine conviction
- [Dialogue TBD]

SHOT 4 (3s): CTA
- Direct to camera
- Eye contact, warm smile
- Expression: recommendation
- [Dialogue TBD]

Notice: no dialogue yet. Just pure visual design. The script comes next, fitted to these durations.

The AI generates all 4 shots with smooth transitions. Same character throughout.

15 seconds of UGC in one generation.

The 2-of-4 Rule: At least 2 of your 4 shots must have a meaningful position or angle change. If every shot is "sitting at counter, talking" → transitions feel weird and there's no visual progression.

Build variety into every storyboard:

• Shot 1: Picks up phone, starts talking

• Shot 2: Leans in, holds up product (closer framing)

• Shot 3: Stands up, demonstrates product

• Shot 4: Direct to camera, different angle

STEP 3: Write the Script to Fit

Now that your visual storyboard is locked, write dialogue that fits each shot's duration.

The word budget: ~2.5 words per second. A 3-second shot gets ~7 words. A 5-second shot gets ~12. Go over and the lip-sync feels rushed. Go under and there's dead air.

Generic AI copy = obvious fake. Here's what actually sounds human:

The Testimonial Script (15 sec)

"Okay I need to talk about [product]. I've been using it for
[specific timeframe] and honestly? [Specific result]. Before
this I was [previous struggle]. Now I just [positive outcome]
every single time. If you're dealing with [pain point], you
need to try this."

What makes it work:

• "Okay" and "honestly?" are filler words real people use

• Specific timeframe (not "a while" but "three weeks")

• Before/after structure

• Direct recommendation at end

The Unboxing Script (15 sec)

"Look what just came in. I've been waiting for this [product].
[Opening reaction]. Oh wow okay, [first impression about
quality/design]. I'm going to try it for [timeframe] and update
you guys. But honestly [initial positive signal]. This is gonna
be good."

The Problem/Solution Script (15 sec)

"I used to [specific struggle] all the time. It was so [how it
made them feel]. Then I tried [product] and [what changed]. Now
I [positive result] without [previous hassle]. If you deal with
[pain point], this is literally the answer."

The pattern: Specific details + filler words + natural rhythm. Map every line to a shot from your beat sheet. The script serves the visuals, not the other way around.

THE PROMPTING SECRET (Think Shots, Not Keywords)

Most AI-generated UGC gets rejected by viewers in the first 2 seconds. Human brains detect unnatural movement.

We've spent our entire lives watching real people move. Our pattern recognition is insanely sophisticated. When movement is wrong, even slightly, we feel it immediately.

"Something's weird about this."
"This doesn't look right."
"Fake."

The issue: Most people prompt AI video like they prompt images. They describe what's in frame, not how it moves.

"Woman talking about skincare."

Kling generates motion. Your prompt needs to describe movement patterns that match what humans unconsciously expect.

Here's what human brains expect from real UGC:

• Slight handheld drift (phones aren't tripods)

• Natural weight shifts (people don't sit perfectly still)

• Organic gestures (we talk with our hands)

• Environmental interaction (touching products, moving objects)

• Imperfect framing (real people don't compose perfect shots)

What makes human brains reject it:

• Perfect stability (this was on a tripod = production)

• Unnatural movements (AI physics that don't match reality)

• Zero camera drift (phones always have micro-movement)

• Robotic gestures (programmed, not natural)

The fix is prompting for the imperfections humans expect.

The 6 Elements of a Strong Prompt:

The 4 Rules:

1. Motion verbs matter: "dolly push-in" not "camera moves closer"

2. Texture = credibility: condensation, fabric sheen, visible breath

3. Describe temporal flow: beginning → middle → end in one shot

4. Name real light sources: "morning window light" not "nice lighting"

Weak vs Strong:

Write prompts like single-shot film directions. One flowing sentence, not a keyword list.

THE HOOK LAYER (This Is Where Most AI UGC Fails)

Your hook needs to happen in the first second. Not after buildup.

Most people focus on the script. But UGC hooks work in layers:

The Best performing UGC uses 2-3 layers simultaneously.

Here's the hook formula that actually works:

Discovery Hooks (High Performers):

"Sooo apparently I've been [doing X] WRONG"
"Been [doing X] for [time] and NOW I find this?"
"I could LITERALLY KISS the person that showed me this"

Disbelief Hooks:

"Wait... did that actually just work?"
"I'm not even kidding right now"
"Okay I need to talk about this"

Pattern Interrupt Hooks:

"Stop scrolling. This is important."
"If you [pain point], you need to see this"
"I've been gatekeeping this for too long"

The pattern: Audio hook + visual reaction + optional text overlay.

Your AI character says the hook while their face shows genuine surprise or excitement. Text reinforces with different words. Three layers, one second.

THE AUDIO TRAP (Your Brain Knows Something's Wrong Before You Do)

Here's what nobody talks about: your audience has been trained since
birth to detect fake human behavior.

Not consciously. Unconsciously.

When audio is too clean, too perfect, too "produced", people don't
think "that's an AI ad." They just feel something's off. And they scroll.

I tested this with a client last week. I generated 10 testimonial
videos. Clean audio on 5, imperfect on the other 5.

The clean videos: 0.6% CTR
The audio: 2.8% CTR

People just didn't engage.

Here's what authentic UGC audio actually sounds like (and why your
brain trusts it):

• Room reverb (your brain expects enclosed spaces to have echo)

• Faint background ambience (real life has ambient sound)

• Casual cadence with natural pauses (nobody speaks in perfect sentences)

• Slight volume variation (phone mics aren't studio equipment)

• "Um"s and self-corrections (thinking while talking is human)

The algorithm doesn't judge. But your audience's unconscious pattern
recognition? That's been training for 200,000 years.

THE "STEALTH CHECK" (Does It Smell Like An Ad?)

Here's the uncomfortable truth: most AI-generated UGC fails not because the video quality is bad, but because it smells like an ad.

Before generating, I run every concept through 4 mandatory checks. They're about matching what humans unconsciously expect from authentic content.

The 4 HUMAN DETECTOR CHECKS:

1. CAMOUFLAGE (0.5s Test): Would a human scroller identify this as "someone's real content" or "a brand's ad" in the first half-second?

   What makes humans scroll:
   • Lighting too perfect (our brains know what phone cameras look like)
   • Fonts look corporate (we've seen 10,000 ads, we know the patterns)
   • Audio is stock music (triggers "commercial" association instantly)

   The fix: Make the first frame look like a person filmed it on their
   phone. Because that's what we're claiming it is.
   

2. VIBE (3s Test): Did the viewer stop scrolling but leave within 3 seconds? That's the "something's off" signal. They gave it a chance. Their unconscious
   mind rejected it.
   
   What triggers rejection:
   • Story is boring (no reason to keep watching)
   • Setup takes too long (real UGC gets to the point)
   • Creator sounds scripted (our brains detect rehearsed speech patterns)
   
   The fix: Lead with value (entertainment or education) before any
   product mention. Earn the attention.
   

3. INTEGRATION (when the product appears): When the product shows up, does the viewer swipe away? That's the moment humans detect "oh, this is a sales pitch."
   

   • What makes humans bail:
     • Transition from content to product is clumsy
     • Tone shifts to "sales voice" (we've heard this voice 1000 times)
     • The product appearance feels forced, not organic
     

     The fix: Make the product appearance inevitable. A natural detail in
     a genuine moment, not a scripted reveal.
     

4. IMPERFECTION (Authenticity Audit): Does this feel like a real person's content or a marketing team's output?
   
   Humans know the difference, even if they can't explain it.
   

   • What triggers the "fake" detector:
     • Too polished (real people don't have production teams)
     • Generic scripts ("This product is amazing" - nobody talks like this)
     • Perfect delivery (real humans pause, restart, say "um")
     

     The fix: Add deliberate imperfection. Slightly off framing. Natural
     pauses. Casual language ("Honestly?" "Okay so" "You guys").

If any test fails humans scroll. Meta sees low engagement. The ad dies.

You're not gaming the system. You're creating content that passes the
test humans have been running unconsciously since childhood.

STEP 4: The Variant Multiplier

Now we’re going to flip the economics.

At $500 per real UGC video, that's $7,500 in creative testing.

With this system: maybe $50 in API costs.

WHAT I'M ACTUALLY USING THIS FOR

Client UGC at Scale

We run paid social for DTC brands. UGC is always the bottleneck.

We're not replacing human UGC creators. We're filling the gap between "we need 10 creative variants" and "our budget is 2 videos."

Testing Before Investing

Here's the smart play: AI-generated UGC as creative testing before paying real creators.

Generate 10 AI testimonials with different hooks, angles, demographics. Run them at low spend. See what resonates.

The winner? Now brief a real creator to recreate that winning combo.

You're not guessing anymore. You're validating with data before writing the bigger check.

THE HONEST LIMITATIONS

Is this perfect? No. Here's what doesn't work yet:

Lip Sync Drift: Once you hit 10 seconds the lip sync starts to drift. Ask the skill to generate in smaller chunks if you find this.

Zero editing: You'll still want to add captions, music, and a cut or two. This isn't "press button, receive finished ad."

Generation artifacts: Sometimes you get weird stuff, object morphing (phone turns into product), extra hands appearing, face drift mid-video. It's stochastic failure. Just reroll. Expect around 4 regenerations per final video.

Scene transitions: If your shots are too similar (sitting → still sitting → still sitting), transitions feel awkward. Build visual variety into your storyboard. Position changes, angle changes, framing changes.

But for the core job, consistent character + natural testimonial + multiple variants, it's shockingly good.

HERE’S THE MATH

This definitely DOES NOT replace your best-performing human creators. What it replaces is the $15,000/month you're spending on "good enough" volume.

THE BIGGER PICTURE

Six months ago, AI video was a novelty, unreliable. Interesting demos, unusable for real work.

But that gap between "cool tech demo" and "wait this might be actually useful for my business" is closing fast.

But now that we have character persistence, product consistency, and realistic motion, that changes the calculus. The power here isn’t creating one-off videos. Think of this running across dozens of stealth accounts, finding the best hooks, angles, strategies all while getting eyeballs, and then giving your human creators the best of the best. That's a content strategy.

The brands that figure this out first will have 10x the creative testing velocity at 1/10th the cost.

The UGC arbitrage window is open. It won't stay open forever.

Get the Skill (Free)

I'm giving away the entire UGC Content Factory skill. The same system I use above, automated.

DOWNLOAD THE FULL SKILL

You get:

• The complete Claude Code skill (drop in and run /ugc-content-factory)

• Full director-first pipeline: Creative Director → Cinematographer → Screenwriter → Engineer

• Character library with archetypes and saved character templates

• Scene settings reference

• Kling 3.0 prompt templates and API integration

• Just add your fal.ai API key and go

Takes about 10 minutes to set up. Then you're generating.

Want Us to Build Your UGC Army For You?

If you'd rather skip the learning curve and just get the content, we're opening a few UGC strategy sessions this month.

We'll build your first character roster, generate test videos for your product live, and map out a variant testing plan so you know exactly which hooks and creators to scale.

Book a UGC Strategy Session →

Reply and tell me: what product would you test first with this system? I read every response.

Let's build,

Matt

P.S. The irony isn't lost on me that I spent years building relationships with great UGC creators... and now I'm teaching you to replace some of that work with AI. The difference? The best human creators are still irreplaceable. What's replaceable is the $400 "good enough" middle tier. AI just became the new floor.

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

The AI Workforce Works While You Sleep

The AI agent that smashed Cloudflare's stock ticker this week?

100k GitHub stars. Fortune. Forbes. Wikipedia article.

I'm now a power user.

It's called OpenClaw now, after Anthropic's lawyers got involved with the creator (more on that in a second). But the point isn't the hype.

The point is what's inside: a system that gives you 100s of employees who never sleep, never complain, and cost pennies per hour.

Twitter tweet

Not hypothetically. Right now. On your laptop. Hell, I even run mine from my phone.

Here's the exact setup.

Become the go-to AI expert in 30 days

AI keeps coming up at work, but you still don't get it?

That's exactly why 1M+ professionals working at Google, Meta, and OpenAI read Superhuman AI daily.

Here's what you get:

• Daily AI news that matters for your career - Filtered from 1000s of sources so you know what affects your industry.

• Step-by-step tutorials you can use immediately - Real prompts and workflows that solve actual business problems.

• New AI tools tested and reviewed - We try everything to deliver tools that drive real results.

• All in just 3 minutes a day

Join 1M+ pros

THE PROBLEM: You're the Bottleneck

I've built agencies, scaled brands (you know the Fireball Whisky story), and run multiple projects at once. The bottleneck is always the same: me. (I wrote about this exact problem in I Automated the Word 'No')

Not talent. Not capital. Not even just time. It's attention.

Every minute I spend on routine tasks (research, drafting, reviews, scheduling, data pulling) is a minute I'm not spending on the things only I can do: strategy, relationships, creative direction.

Most founders try to solve this with:

• Hiring (expensive, slow, management overhead)

• Outsourcing (quality control nightmare)

• Automation tools (fragmented, don't talk to each other)

But here's what changed everything for me: AI agents that actually work together.

MY ACTUAL SETUP: Meet the Team

For the last week, I've been running a multi-agent framework called Clawdbot (now open-sourced as OpenClaw) that coordinates multiple AI agents, each with their own specialty.

Think of it like a company org chart (except every employee is an AI).

They coordinate via a NocoDB dashboard called "Squad HQ", agents can see each other's tasks, leave notes, and hand off work.

The brutal truth? This setup replaces what used to require 5-10 juniors for routine work.

THE WEEK IT BROKE THE INTERNET

The creator is Peter Steinberger. He built Clawdbot as a side project after he sold PSPDFKit for $119 million. About a week ago Clawdbot went viral on X.

Within 72 hours:

• Anthropic's lawyers sent a cease & desist (can't use "Claude" in the name)

• Crypto scammers launched a fake $CLAWD token (hit $16M market cap before crashing)

• Someone hacked his GitHub temporarily

• He renamed it "Moltbot" (like a lobster molting)

• Then renamed it AGAIN to OpenClaw

Twitter tweet

100k GitHub stars. 2M visitors in a week. Fortune, Forbes, CNET, IBM all wrote about it.

Twitter tweet

Cloudflare stock moved 14% because the project uses their infrastructure.

None of this changes what's actually useful: the architecture inside.

THE ARCHITECTURE: How Agents Work Together

Here's what makes this different from just "using ChatGPT":

1. Persistent Memory

Each agent has daily memory files and long-term context. When I talk to Big Players on Thursday about newsletter ideas, it remembers that conversation on Monday. No more "starting from scratch" every time.

Here's the part that blew my mind: the agents maintain their own memory.

Each day, they log what happened. Every night, Jean Luc reviews those logs and updates long-term memory, distilling raw notes into curated context.

It's like journaling, but the AI does it for itself.

When I come back to a project after two weeks, the agent remembers where we left off. The context isn't lost. The decisions are documented. The lessons are recorded.

2. Delegation Hierarchy

Jean Luc (Chief of Staff) can route tasks to specialists. If I ask about the newsletter, it goes to Big Players. Client question? Emerald Ops. Product feedback? StealAds.

3. Agents That Spawn Agents

Here's where it gets wild: specialists can spawn sub-agents for complex tasks.

Working on this newsletter right now? A sub-agent researched my previous newsletters for voice and tone while another pulled context on my current setup. They complete their work, report back, and disappear.

Think about that: An agent creating temporary agents to complete tasks, then synthesizing the results.

4. Proactive Check-Ins (Heartbeats)

Every 30 minutes, Jean Luc gets a "heartbeat", basically a wake up call to check if anything needs attention.

Here's what that actually looks like for someone running a brand:

For agencies/service businesses:

• Ad spend pacing 40% over budget? You know before they do.

• That campaign you launched yesterday tanking? Alert hits your phone at 7am, not when the client calls at 2pm.

• Competitor just dropped a new campaign in Meta Ad Library? Brief lands before your morning standup.

For growth marketers:

• Brand mention spiking on social (good or bad)? Context before the CEO asks.

• Creative team missed a deadline in Asana? Flagged before the launch date slips.

• That influencer partnership going sideways? Early warning, not damage control.

The difference: you're operating from awareness, not constantly checking dashboards and praying nothing's on fire.

If something's worth flagging, Jean Luc messages me. If not, silence. No notification fatigue. Just the stuff that actually matters.

5. Real-World Connections

These agents aren't just writing text. They're plugged into the systems where work actually happens.

For marketing/agency work, mine can:

• Pull competitor creative from Meta Ad Library and break down what's working

• Monitor brand mentions across platforms and flag sentiment shifts

• Update client dashboards with live campaign data (no more manual screenshot hell)

• Draft creative briefs based on what's actually performing, not gut feel

• Cross-reference campaign performance against industry benchmarks

• Track share-of-voice changes week over week

The infrastructure:

• n8n workflows connecting everything

• MCP servers for database/API access (Google Analytics, Meta, LinkedIn, you name it)

• Browser automation for the stuff that doesn't have an API

• Messaging integration so agents can ping me on Telegram, Slack, wherever I am

This is the power of MCP (Model Context Protocol) I wrote about a few weeks ago (USB-C for AI connections).

The unlock: When your AI can actually pull the data, analyze it, and draft the recommendation (instead of just suggesting you go do that) you're saving the mental overhead of remembering to do it at all.

THE PART THAT CHANGED EVERYTHING: OVERNIGHT MODE

Here's what happens when I go to sleep:

Jean Luc doesn't shut down. He checks if there's proactive work to do.

StealAds Builder pushes code. Emerald Ops drafts client updates. Big Players preps tomorrow's content.

I wake up to a summary: what got done, what needs my attention, what's blocked.

No morning standup. No inbox blowing up. Just: "Here's what happened while you were sleeping"

It's shipping.

KILLER USE CASES

Here's where it gets interesting. Not the generic "AI can help with content!" stuff. The specific things that made me go "oh, this changes how I operate."

The "Why Is This Campaign Underperforming?" Autopsy
Client calls asking why Meta ROAS dropped 30% this week. (This is the same workflow I covered in How I Cut Meta Ads Reporting to 45 Seconds) Old way: scramble, pull reports, build a deck, hope you find the answer before the call.

New way: Emerald Ops already ran the analysis overnight. Creative fatigue on the top 3 ads. CPMs up 22% in their core demo. Competitor increased spend in the same auction. Here's a draft response with three recommended actions.

I show up to the call looking like I had 20 senior analysts working overnight.

Competitive Intel That's Actually Current
Not a quarterly report that's stale. I'm talking: "Competitor X just launched a new landing page with different positioning, here's what changed and what it might mean for our messaging."

The agent monitors their site, their ads, their social. When something shifts, I know. Not because I remembered to check. Because it told me.

Brand Voice at Scale (Without Brand Drift)
Here's a problem nobody talks about: the more content you produce, the more your brand voice fragments. Different writers, different contexts, different platforms and entropy kicks in.

My content agents have my last 50 newsletters, my voice guidelines, my "never say this" list. When Big Players drafts something, it's not generic AI slop. It's my patterns, my cadence, my opinions.

I still edit. But I'm editing from "pretty close" instead of "this sounds like a robot trying to be me."

The Monday Morning Deck Problem
You know the one. An investor, client, even your boss wants a strategic recommendation by Monday. It's Friday at 4pm. You need research, competitive context, data analysis, and a coherent narrative.

Old way: weekend ruined.

New way: I brief Jean Luc on what we need. Overnight, sub-agents pull competitive intel, analyze our campaign data, draft the strategic options, and compile it into a structured brief. I wake up Saturday to a first draft.

I still do the strategic thinking (you ARE NOT outsourcing your brain, if you do that you're screwed). But the assembly, the part that used to eat weekends, happens while I sleep.

Post-Mortems That Actually Get Written
Every campaign ends with good intentions: "We should document what worked."

Three months later: nothing documented, lessons forgotten, same mistakes repeated.

Now? When a campaign ends, Emerald Ops automatically compiles: what we tested, what performed, what we'd do differently, key learnings for next time. Stored, searchable, actually useful. Our next project is built automatically with these lessons baked in.

Institutional knowledge that doesn't walk out the door when someone quits.

HOW TO THINK ABOUT THIS FOR YOUR BUSINESS

After 20 years of scaling marketing operations (including taking Fireball to a billion-dollar brand) I've learned one thing: the constraint is never effort. It's attention.

You can hire more people. You can work more hours. But you can't manufacture more of you paying attention to the thing that matters right now.

Here's how I'd think about this if I were starting over:

The Quality-at-Scale Problem

When we scaled Fireball, the hardest part wasn't the strategy. It was maintaining quality as volume exploded. Every new market, every new campaign, every new channel there was more surface area for things to go wrong.

The playbook that worked: systematize the repeatable, protect time for the irreplaceable.

Same principle applies here. AI agents handle the repeatable like research, drafts, monitoring, reporting. That protects your time for the stuff that actually requires your judgment: strategy, relationships, creative direction.

It's about doing the right things, not just more things.

The "Hire Slow, Fire Fast" Parallel

You know the hiring advice: don't just fill a seat, find the right person for a specific role with clear accountability.

Same applies to agents. Don't just spin up "an AI assistant." Build a specialist:

• Clear domain → "You own competitive intelligence" not "help me with stuff"

• Defined outputs → "Weekly competitor brief, anomaly alerts, quarterly deep-dives"

• Context that compounds → Past briefs, your strategic priorities, what matters to your business

• Boundaries → What they own, what they escalate, what they never touch

The agents that work are the ones with jobs.

Start With One Workflow, Not One Tool

Don't start by asking "what AI tool should I use?"

Start by asking: "What's the workflow that's killing me?"

For most brand leaders it's one of:

• Reporting → Data exists in 5 places, someone spends 4 hours assembling it weekly

• Competitive monitoring → You know you should track competitors, you never actually do

• Content production → You need 10x the output, you have the same team

• Comms → Status updates, QBRs, "just checking in" emails that eat hours

Pick ONE. Build an agent that owns it end-to-end. See what changes.

Then stack.

The Integration Tax

Here's what separates "using ChatGPT" from "building an AI workforce":

ChatGPT is a chat window. Useful, but isolated.

An AI workforce is connected. It connects to your data, your tools, your communication channels. When the agent can pull this week's campaign performance from Google Analytics, compare it to last week, draft an analysis, and send it to Slack without you touching anything... that's the magic moment.

The integration work isn't sexy. Setting up CLIs, MCP servers, connecting APIs, building workflows in n8n. But it's where the leverage lives.

If your AI can only suggest things, you're saving minutes. If your AI can do things, you're saving hours. And more importantly, you're saving the mental overhead of remembering to do it at all.

THE UNCOMFORTABLE TRUTH

This isn't about replacing people. Far from it. My best people are irreplaceable and always will be. And my best people each with 100 agents in their pocket are an unstoppable force.

This is IS about:

• Removing the operational tax that drains founders

• Scaling attention across multiple priorities

• Maintaining context that would otherwise get lost

• Moving faster without proportionally growing headcount

The companies that figure this out will operate at 10x the pace of those that don't.

The gap between "using ChatGPT" and "building an AI workforce" is about to become a competitive moat.

WHAT I'M BUILDING TOWARD

Here's the vision: a future where every small team operates like a company 10x their size.

Because AI handles the 80% of ops work that used to require headcount.

The founder focuses on strategy and relationships. The AI team handles execution.

We're not there yet. But we're closer than most people realize.

If you're running a business and not thinking about this, you're going to wake up one day wondering how your competitor is shipping so much faster with a smaller team.

The answer will be: they have 100 employees in their pocket.

GET THE SYSTEM

OpenClaw is open source. That's the free resource.

→ Get OpenClaw on GitHub

100k developers already starred it this week. The docs, the architecture, the agent templates, it's all there.

If you want to see what AI employees are doing for GTM at scale, give me shout 👇

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}

Riley Brown spent a few grand a few months back on his startup's launch video. Last week he made a better one in 30 minutes.

Riley’s the co-founder of Vibecode app. They do over $100K MRR. They raised $9 million, Alexis Ohanian is on the cap table.

These aren't people who can't afford good video. They paid good money for their launch video. Video studio. Rounds of revisions. The whole dance.

Twitter tweet

Last week, Riley sat down with Remotion Skills and made a better one in 30 minutes.

Not "comparable." Better.

Twitter tweet

And the video editors I showed? They aren't scared. They started using it that night.

Turn AI into Your Income Engine

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

• A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential

• Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background

• Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

What Remotion Actually Is

Twitter tweet

Remotion is a framework that lets you build video programmatically, code that renders as video. Remotion Skills is the AI layer on top: natural language to branded motion graphics.

You describe what you want. It builds it. Your brand colors, your typography, your movement language.

The Skill Gap Moved.

Here's what those video editors I spoke to understood instantly (and what the video agencies charging $50K/month retainers are hoping you won't figure out):

Video editing has two layers.

Layer one is taste. Timing. Storytelling. Context. Client intent. Taking raw footage and making someone feel something. That's hard to encode. That's human. That's what great editors actually do.

Layer two is production. Lower thirds. Intro sequences. Data visualizations. Product showcases. Branded transitions. Social templates.

That's repetitive. That's template-driven. And as of last week, that's repeatable.

The $10K/month video editor isn't getting fired.

The people who should be terrified aren't the skilled editors. It's the Canva power users. The template marketplace. The production houses charging premium prices for commodity work while hiding behind "creative process."

The $10K budget didn't disappear. It moved. Production costs collapsed. What you're paying for now is the eye, knowing what good looks like.

The editors who make $15K/month? They just got a 10x multiplier. Instead of spending six hours building one lower-third animation, they describe it in a sentence and spend those six hours on the storytelling that clients actually pay premium for.

Riley didn't make a better video because he suddenly learned After Effects. He made it because the tool eliminated the gap between knowing what good looks like and seeing it rendered.

What's Actually Happening

This isn't just Riley.

A creator named @agibjames built a brand video for Postbridge (@jackfriks' solo startup) using a workflow that shouldn't exist yet.

Twitter tweet

He grabbed an Apple product video as reference. Fed it to the AI. Had it decompose the motion language i.e. the framing, the rhythm, the transition timing, the way elements enter and exit frame.

Then he swapped in Postbridge's assets and rebuilt it.

Twitter tweet

The result looks like it came from a studio.

Start with a reference you love. Have AI understand it. Rebuild it as yours.

Twitter tweet

That's a repeatable system. And the Remotion announcement hit the creator community a few days ago. YouTube editors saw it. They're already implementing.

I've been paying for video production for over a decade. I've written those checks. I know what $3K gets you and what $50K/month agencies deliver.

What these guys are making looks better. More polished. More on-brand. And they're generating dozens of variations.

What I Built This Weekend

II had to try this myself.

I spent about 1.5 Finchers on it this weekend (that's one viewing of Zodiac plus a little overtime. I measure creative experiments in David Fincher films now).

Here's my process:

Step 1. Found a reference I loved.

I downloaded an Instagram video from Greg Isenberg (his short-form motion graphics style is clean, punchy, and recognizable).

Instagram post

2. Decomposed it with AI.

I knew Gemini was strong at multimodal analysis.

So I ran the video through a series of prompts: break down the scene composition, the animation timing, the color palette, the typography treatment, the transition styles. Everything.

3. Rebuilt it in Remotion.

Took those decomposed elements and generated my own components. My own colors. My own brand language.

The result isn't polished yet. But I now have animation styles, component structures, motion patterns that compound. Every component I add makes the next one faster.

That's the part that matters.

I built the foundation of a motion identity system.

The Branded Motion Library: Why This Is Actually a Moat

Most brands think in static terms. Logo. Colors. Fonts. A brand guidelines PDF that nobody reads and nobody follows consistently.

But what if your brand identity wasn't a document?

What if it was a system that produces?

That's what this means. Motion identity, encoded and executable.

Twitter tweet

The Postbridge example shows the exact architecture:

1. Reference capture. Start with motion you love. Doesn't have to be in your industry (@agibjames used an Apple ad for a SaaS product).

2. Decomposition. AI breaks down the movement language. Not "what happens" but "how things move." The velocity curves. The easing. The rhythm between elements.

3. Reconstruction. Rebuild with your assets. Same cinematic grammar, but for your brand vocabulary.

Now you have a reusable component. A system primitive.

Twitter tweet

Build five of these and you have a motion library:

• Intro sequence: How your brand enters frame

• Outro/CTA: How you close and drive action

• Lower thirds: How you present information

• Data visualization: How you display numbers

• Product showcase: How you present what you sell

Each one encodes your taste. Your movement philosophy. Smooth and elegant? Punchy and aggressive? Minimal and restrained?

Once built, you prompt infinite variations. Different content, same identity.

Build once, render infinitely.

Every asset you create feeds the next one. Your motion library becomes a competitive advantage (not because nobody else can build one), but because yours is already six months deep while they're scheduling their first "creative discovery call."

The Bottom Line

The window is 12 months.

I’m not waiting for this to be polished. I’m compounding while everyone else evaluates.

This is what happens when taste becomes scalable.

This is what happens when brand guides become executable.

The skill gap moved from knowing how to push pixels to knowing what good looks like. That’s now the expensive part.

The brands/creators going to win in 2026 won’t be making one-off videos. They’ll be making video systems.

— Matthew Berman

If this newsletter lit a fire under you, forward it to one person who would benefit.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do. 

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}