I used to write every commit message by hand, now I let an AI draft them and only edit the bad ones
Two years ago I would sit there for 5 minutes trying to word a commit message right, and half the time it still said something useless like "fix stuff." Now I paste the diff into a model, it spits out a draft in about 3 seconds, and I rewrite maybe 1 out of 5 because the rest are fine. My rule is simple: the AI drafts, I approve, and I still read every single one before pushing. It cut my commit time to almost nothing, but I do not trust it on anything that touches auth or billing because it gets way too confident there. Last month it called a small CSS fix a "critical security patch" and I caught it right before it hit main. Curious if other people let the model commit straight to the repo or do you still gate it like I do?