I will not build dev tools

Three times now I've started a product for developers. Three times it got commoditized before I finished it — because developers build their own version in an evening, and whatever they don't build, the model companies ship as a feature by next quarter.

I will not build dev tools. I will not build dev tools. I will not build dev tools.

Three things that got commoditized faster than I could finish them

Here's what I mean, concretely. Over the past year I started three products, all aimed at people like me — developers who use AI agents all day. A code editor, because I wanted mine to work differently. An app for using Claude from my phone, because the ideas show up when I'm away from the desk. An app for running Claude on a cron, so work happens overnight.

All three still sit in my git history. All three are now buttons inside somebody else's product.

None of them died because they were hard. Each one worked within about a week. They died because by the time I reached the boring last 20%, the thing I was building had turned into a checkbox in something bigger. The gap between "nobody has this" and "everybody has this" is now measured in months.

The tell: I have never once looked at the price

If dev tools were a market I understood, I'd be a customer in it. I'm not. When I see a tool I want, my reflex isn't to buy it — it's to open an agent and have my own by the end of the night.

Not a humblebrag — a warning sign. It genuinely does not occur to me to check what one costs.

So who exactly was I planning to sell to? Somebody with the same reflex I have. They watch the demo, they open their agent, they have it by dinner. I am the target customer, and the target customer doesn't buy. He clones.

The loudest room is not a market

AI dev tooling gets more attention per user than anything else in software right now. That attention is exactly what fooled me. Attention and willingness to pay have come apart, and in this corner they've come apart badly.

My read, not a dataset. The point is the direction of the arrow, not the coordinates.

Developers who live inside AI are the one audience that can rebuild your product the week they discover it — and who enjoy doing it. Highest opinion volume, lowest chance of a card ever getting entered. The trap is baited with applause.

“But I'm only building it for me”

This is the excuse I fall back on when the market argument lands, and it's the weakest one I have. What actually happens is a loop: I want to ship faster, so I build a tool to ship faster, and then the tool is the thing that needs shipping.

Three weekends sharpening an axe I then never carry to a tree.

And even when the tool is good, it expires. Whatever takes me three weekends today is one prompt away in six months. Personal dev tooling has a shelf life measured in model releases.

So here's the filter I use now

Before I write a line of code, I picture the specific person I'd be building for and ask one question about them.

One question, asked before any code. It kills almost everything I get excited about on a Saturday.

There is an enormous amount of opportunity in AI right now. Almost none of it is in building for the people who are already best at using AI. It's out where they aren't looking — people with a real problem, no interest in building anything, and a genuine willingness to pay to make it go away.

If you've caught yourself in the same loop, I'd love to hear what you moved on to. And if I post a code editor next month, please yell at me.