AI Is Not Your Moat

AI tools are taking us to a place where zero mental effort is possible.

If you write, you know the pain of coming up with a clever cover image for your articles. Today you open ChatGPT or Claude and hand the problem to an AI agent. Five minutes.

A task that used to require creativity, design knowledge, and good taste. Anyone can do it now.

So let’s think bigger.

Instead of a cover image, say you want a comic. What’s the difference? Instead of a one-shot prompt (“draw two cats playing the piano”), you need multiple turns: write the script, break it into panels, write the dialogue, render the page.

I did that. A full LLM-powered comic strip factory that runs without supervision.

And if I can do it, anyone can do it too. Eventually.

The Magic Trick

So how do you turn a multi-turn ChatGPT conversation into an autonomous workflow?

You might think it’s about taking notes, devising a playbook, running a few trials until you get consistent results. That’s part of the trick. It’s not enough.

It’s not enough because anyone can do that too. And if anyone can do it, chances are your results are not as valuable as you think.

The Magic Trick is the data. The specific data you inject that nobody else has.

Mine was another pipeline feeding it ideas, plus the feedback from every bad comic that came before.

Every failure, written down as a rule the pipeline checks itself against on the next pass. “No joke the camera can’t photograph.” “Stay in the story’s own domain.”

One failure, one rule. Logged to a ledger the pipeline reads before it runs again. A few hundred of them by now.

That ledger is the whole thing.

Every model inside is rented. Claude does the thinking, gpt-image does the drawing. The same ones anyone can call. I even route them by step: the expensive model where the joke is born, the cheap one everywhere else.

When a sharper model ships, I swap it in. It breaks a few things (a new model finds new ways to ruin a joke), the ledger catches them, I add a rule or two, and the pipeline comes out better than it was. I never rebuild it.

The value was never the model. It was the ledger I’d fine-tuned around it. Models come and go. The ledger survives them all.

And the hardest lesson: more rules never made it funnier. The right ones did.

That comic pipeline is a toy version of the only thing that defends a company now. Almost nobody is pricing it.

What your growth curve used to prove

Growth was never the thing investors were buying from you. It was a proxy.

Build a product people want, get it to grow, and you’ve proved the scarce thing: you can build. Building took a team. Years. Money. That difficulty was your moat, sitting quietly under the curve. Nobody put it on a slide because nobody had to.

That difficulty is gone.

I did the cost math in The $99 Savings Trap: an agent does a hundred-dollar task for about a dollar. Same collapse, your side of the table. If anyone can build your product in a weekend, “we built it and it grew” proves a head start.

Head starts get erased.

So the curve still goes up. It’s just measuring the part that stopped being scarce.

The Winner’s Protocol

Same models, same agents, same tutorials on YouTube. Everyone has them. The difference between a moat and a feature is whether anything accumulates.

There’s an order to it.

Where’s your data? Scattered across chats nobody reads, or in one place the AI can run on? What can that data do for you now, not just tell you? And what runs itself, leaving a receipt every time it acts?

Get the order right and each turn feeds the next. Data feeds the tasks. The tasks become workflows. The workflows throw off receipts (what worked, what got corrected), and those become data again, sharper every loop.

My comic pipeline is that loop in miniature. Every bad panel left a receipt in the ledger. That ledger is what a competitor can’t download.

I will repeat that because it is important: competitors cannot download your ledger.

They could come up with your ledger, but they will have to catch up first.

That’s the moat. Not the model. What the AI makes from what you’ve accumulated, again and again, that nobody can copy in a weekend.

I told the build side of this in Stop Building AI Agents. Here’s the investor version: the model is rented, the agents are a weekend of work, the tools are commodity wrappers. None of it is a moat.

The moat is the loop compounding behind them.

The bar got a second edge

For a while, “can they stand behind it” was a soft question. It hardened.

On August 2, 2026, the EU’s enforcement power over general-purpose AI providers switches on. Fines up to €15 million or 3% of global turnover (EU AI Act, Article 101).

Precisely: not new rules. The obligations have been in force since August 2, 2025. What turns on now is enforcement.

And when everyone said “the AI Act got delayed,” half true. The high-risk rules slipped to the end of 2027. The provider obligations held.

I’m not telling you fines are coming. I’m telling you accountability stopped being optional. A company built on a loop it owns can stand behind its output. A wrapper that ships the exhaust and keeps nothing can’t.

Legal defensibility isn’t a wall you stop at. It’s a thing the moat already hands you.

The ruler goes public

One more clock.

Anthropic already filed its draft S-1, confidentially, with the SEC (Anthropic). When that prospectus goes public, the market prices intelligence at scale in the open for the first time. And every investor reads the same document. One written to justify a number, not to teach you how to judge one.

So before that ruler sets, change the question you ask.

Your Next Move

The wrong question: “How fast are they growing?”

The right question: “What are they accumulating that no one else has?”

Ask to see the ledger. If every AI action vanishes the second it’s done, nothing is compounding. You’re funding a feature. If the receipts stack up and feed back, you’re watching a moat being poured.

I wrote the long version in Founders and Investors Will Be Replaced by AI.

To make the point the hard way, the book has to pass its own test: it’s a living book, rebuilt by a pipeline of agents that logs every change to a ledger and reads it before the next pass. The same loop as the comic pipeline, pointed at prose. Free to read.

The build barrier fell. The defend barrier didn’t.

AI got you over the first one. Everyone else too.