I thought AI would let me build everything. Turns out it just lets me fail faster.

I'm a project manager by day. I don't get to write the code at work, so for years my ideas just sat in a notes app, each one too expensive in time and money to ever really try. Then the tools got good, and the gap between “what if” and “something real people can download” basically disappeared. At night, on my own, I could finally find out.

AI did help me build. What I didn't expect was the more useful half: it helps me fail faster, and a lot cheaper. Failing an idea used to cost me months and real money before I even learned it was a bad one. Now it costs a weekend and a few dollars of tokens. That sounds minor. It changed everything.

When being wrong gets cheap, you can afford to be wrong a lot. You take more shots, bury the bad ones quickly, and keep moving. The graveyard on the home page isn't me being modest, it's the method. Every dead project taught me something a live one never would have.

But the real reason I keep doing this is simpler: I love the process. The building, the arguing with myself about what the product should actually be, the thousand small decisions in between. There's a kind of understanding that only shows up once you've shipped the thing and watched it meet real people. You can read about product forever and not feel it. Useful or not, money or not, I don't regret a single one of these.

Most of what I learned is the same lesson in different clothes: AI gets you to shipped, but it doesn't make anyone want the thing, or pay for it. That part never got automated. It's still the hardest, most honest question in the whole process, and it's still entirely on me. Which, it turns out, is the question I should have been asking first.

So this is the loop now. Have an idea, build the smallest real version as fast as I can, put it in front of actual people, and listen for the truth. Some of them live here. Some are buried here. I've made peace with both.