At least currently, the only use pattern that can withstand complex codebases is as advanced speech-to-text, but without the speech if that makes sense. The main issue with that is phrasing things in English is often far more verbose, so without the speech, it's very often faster to just do it by hand.