Git

I used to think the commit was the natural unit of software history. It is clean: a hash, a message, something CI can chew on and a reviewer can revert, cherry-pick, or blame. Then AI coding agents made the gap around the commit impossible to ignore.

Six repositories, one system, too many clone commands # On a typical microservices project I have five or six repositories open at once: a couple of APIs, a frontend, shared infrastructure scripts, sometimes a docs repo. Onboarding a new developer meant a wiki page full of clone commands, and “which commit of the API works with this frontend?” was answered by asking whoever was online.