Originally published on Oli Kahn's Substack . Cross-posting here for the dev.to community. Most posts about “building with AI” are either hype or demos. This one is neither. It’s a field report. Over the past five days, I shipped a small side project end-to-end — research, positioning, copy, visuals, deployment. Not a toy. Something with real traffic goals and an actual audience I care about reaching. I didn’t use a single human freelancer. I used AI, my keyboard, and a lot of coffee. Here’s what I actually learned — not the parts that sound good on Twitter, the parts that cost me hours and forced me to rewrite my mental model. 1. AI is a terrible strategist and a spectacular executor The thing nobody tells you: if you ask an AI “what niche should I go into?” — you’re going to get mush. Generic. Hedged. Three options that all sound plausible and all lose. But if you come with a hypothesis — even a half-formed one — and ask “pressure-test this, what breaks?” — suddenly you get real thinking.…