Pick the Project You Already Know Something About

The first few months I started using Claude Code, I had four projects open every day. All four were sitting at roughly the same place: sixty percent done, none of them shipped, all of them waiting for "later this week."

Building with agents makes starting a project easy. It doesn't make finishing one easy, and it doesn't make picking the right one any easier either.

Here's what actually narrows it down: pick something you have domain expertise in, or something personal to you. Both point at the same thing. If you know the domain, you can tell the difference between an agent's output that's actually good and output that just looks plausible. You can catch the wrong assumption, the missing edge case, the thing that would embarrass you in front of an actual user. If the project is personal, you've probably already spent years thinking about it without realizing it counts as expertise, and you have the kind of motivation that survives a bad week.

Picking the project you already have standing to judge, and staying on it instead of restarting the moment something shinier shows up, is the actual fix.