Buenos Bakery, by hand

The first thing we ever built was a website for a bakery.

Nothing about it was ambitious. A small product for a local business - the kind of thing you would expect one developer to knock out in a couple of weeks. I built it in Windsurf, an AI-powered editor, and for the first time in fifteen years of writing software, I was not really typing the code. I was describing what I wanted, watching the editor write it, reading what came back, and deciding whether it was right.

I have fifteen years in engineering, six of them as an engineering manager. I know what good looks like - not as a feeling, but as a checklist. Linting that actually runs. Tests that mean something. A folder structure you can navigate at 2am. Files small enough that a human - or, it turns out, a machine - can hold one in their head. So when the AI handed me code, I did not just accept it. I reviewed it the way I would review a junior's pull request: line by line, does this belong here, is this the right shape, will this rot in three months.

And it worked. Buenos Bakery got built. The AI was fast, I was careful, and between us the thing shipped. It genuinely felt like magic - the first time the machine wrote something non-trivial and correct, I remember thinking the job had just changed permanently.

Then I tried to do it again. And again. And the magic ran into a wall, and the wall was me.

Here is the problem nobody warns you about. When one AI writes code in one editor, you are the reviewer, the integrator, and the quality gate - all of it, all at once, in a single window looking at a single project. The AI got faster the more I used it. I did not. I was still reading every diff, still holding the whole project in my head, still the one saying yes or no to each change. One window, one project, one human bottleneck.

That is fine if you are focused on one thing at a time. But when you realize the only thing stopping you from moving faster is you, something shifts. You want to deliver more. I did not want to build one app well. I wanted to build the way a team of twenty builds, except the team was me and a growing pile of AI. And the editor-shaped setup I had could not do that, for a reason that took me embarrassingly long to name:

I had made the AI faster at writing code without making myself faster at everything around it.