I've been having mixed results with my AI model on my codebases -- sometimes it finds the obscure bug and fixes it, other times it attempts to rewrite previously correct code into utter garbage, breaking it even more.