I had a project sitting dormant for about ten years. The idea was small enough to fit in a sentence: schedule a merge of a GitHub pull request at a specific time.

I pointed Claude at it and let it rewrite the whole thing in Rails 8. Ruby 3.3, SQLite, Solid Queue holding the job until the merge time, no Redis, no Node build step, one process. I deployed it, started using it, and have barely touched the code since. It just runs.

I use it for two things. My blogs are GitHub-repo-and-Actions-backed, so scheduling a merge is scheduling a post to go live. And when I want to deploy a service, I can time the merge for the quiet hours when the fewest clients are active.

There are surely a dozen ways to do this, and some of them are better. GitHub Actions cron, a scheduled workflow, some SaaS with a slick UI. But the appeal here isn’t features. It’s that the entire tool fits in my head. When it does something, I know exactly why. When I need to change it, there’s nothing to relearn. That kind of trust is hard to get from a tool built to do everything, and it turns out to be worth more to me than any feature I’d be trading it for.

Reviving a dead idea and handing the rewrite to an agent got me there in an afternoon. The best part is how little I think about it now.