· 2 min read
Audit your ai projects, then do something about it
I Ran an Audit on a Year of Projects Built with AI. Then I Didn't Touch the Winner for a Month.
My /Documents folder has lots of projects that I’ve built with AI assistance over the past year. Some finished. Most half-finished.
So I audited it. Properly, in stages, using Claude to do the parts that don’t need my judgment and me to do the parts that do.
Stage one was just an inventory. What exists, what stack it’s in, when I last touched it, whether the git history is even coherent.
Stage two went looking for duplicates I hadn’t noticed I’d made. Not file duplicates — idea duplicates. I’d built “a system to catalog and reuse UX patterns” five separate times over five months.
Stage three was a clearance pass I had to make sure anything I publish under my own name has to be unambiguously mine.
Stage four was a market-reality pass, and I asked for the unflattering version on purpose. For everything that cleared the clearance check, I wanted actual competitor research — GitHub star counts, demand signals, whether a funded company already owned the space. Not “great idea!” enthusiasm. Real numbers.
Stage five scored what survived against a fixed rubric — legibility, whether it had a real demo moment, install friction, whether it could pull an ecosystem around it, audience, timing, whether I was actually the credible person to ship it, and how far it sat from a real v0.1. Not vibes. A number.
One project came out on top. A catalog of UX interaction patterns, but built for a world where AI coding agents are doing a lot of the implementing. Out of that session I had an actual task breakdown. A real before/after demo. A launch plan — a Show HN post, a LinkedIn post, even a pre-written reply to the obvious skeptical comment I could already predict.
Then I didn’t touch it for a month. Time to change that.
