Ah, okay. I guess I haven't been doing the agentic version. I type in commands like, "Fix this merge conflict." or "refactor this function into this function, that function, and this service" or stuff like that.

I'm probably not doing that right, since I micromanage the AI and review every step and constantly intervene. What I've seen so far hasn't made me inclined to let it take over the codebase, unsupervised, tho.

The code it writes keeps improving, but it's increasingly opinionated and breaking my data security measures and then lying about it and stuff. If keeps getting smarter, but also more naughty, and I have to audit more and more carefully.