Non-technical founder here. I built our entire AI platform myself using AI coding tools. AMA about what broke.
I’m a former CMO type. My background is brand marketing, not engineering. Over the past year I built my company’s computer vision platform end to end using AI coding tools, and it’s now in production with real customers and real revenue.
The “anyone can build now” narrative skips the painful parts. I lived all of them, and I’d rather post the ugly version than the highlight reel.
\*\*Things that broke, in order of how much they hurt:\*\*
A memory leak in our video processing queue. Two weeks lost. The AI kept suggesting fixes for symptoms instead of the cause. I finally found it by learning to read heap snapshots myself, which I never planned on doing in this lifetime.
Database migrations. I once let an AI-generated migration run against production. It worked, but only because we got lucky on ordering. I now stage everything and read every line. Learn from my near-miss.
Auth. Do not let AI freestyle your authentication. I ripped ours out and rebuilt on a standard library after realizing the custom version had gaps I only understood after reading about session fixation at 2am.
My own overconfidence. After three months of wins I started skipping the “explain this code to me” step. That is exactly when bugs started shipping.
What I’d tell any non-technical founder trying this: you can absolutely get to a real product. You cannot get to a real product without becoming at least semi-technical along the way. The tools don’t remove the learning, they compress it.
Ask me anything, including costs, timeline, and the stuff I’d never do again.