Building
AI Is Making Execution Cheap. That Changes What a Founder Is For.
A small team can now do things that recently required departments. That's extraordinary. It also means execution itself is becoming less of an advantage. I've been thinking about what becomes valuable when almost everyone can build faster.
Having built an AI company, I'm allergic to most of the conversation about AI and startups. It swings between two equally lazy extremes — that AI will replace everything, or that it changes nothing real. Neither is true. What's true is quieter and more consequential: AI is making execution cheap, and that changes what a founder is actually for.
The real shift isn't replacement, it's leverage
The most honest description of what current AI does for a small team is that it raises the ceiling of what two or three people can produce. It doesn't remove the need for judgment, taste, or domain understanding. If anything, it makes those more valuable — because it lowers the cost of generating options and raises the cost of choosing the right one.
A founding team used to be limited by execution bandwidth: not enough hands to write the copy, draft the model, build the dashboard, test the flow, research the market. AI doesn't do any of those perfectly. It does all of them serviceably, and instantly. The bottleneck moves from producing work to directing it.
Faster loops change the strategy, not just the schedule
What matters most in early building is the speed of the loop: build something small, put it in front of reality, learn, repeat. AI compresses that loop. You can draft, test, and discard in an afternoon what used to take a week.
That isn't a productivity talking point. It's a strategic one. When loops are cheap, you can run more of them, which means you can afford to be honest about killing the ideas that don't work — because the cost of finding out is low. Founders who use AI to do the same plan faster are missing the point. The point is to run a different, more experimental plan, because the cost of being wrong has dropped.
Where the founder still has to show up
Here's where I push back on the replace-everything crowd. AI is genuinely good at the middle of the distribution of a task. It's bad at the edges — the novel constraint, the specific regulatory reality, the thing a particular customer actually needs versus what they say they need. As execution gets cheaper, the founder's job migrates to the edges.
In industrial AI, this was obvious. The model could see patterns across a plant that no operator could hold in their head — that was the leverage. But knowing which pattern mattered, which action was safe, and which "optimization" would break something downstream — that was still a human loop, and a high-stakes one. The same is true in any venture: AI generates, the founder decides.
The new job description
If execution is cheap, what's the founder for? Mostly for the things that were always hard and that cheap execution doesn't touch: choosing the problem, reading the system, holding the standard, and deciding which of the ten things AI can generate is the one worth building. Leverage amplifies direction. Get the direction right and a small team moves like a large one. Get it wrong and you arrive at the wrong place faster than ever.
AI isn't the product. It's the leverage behind the product. The founders who understand the difference are the ones building things that last.