As more LLM generated code lands in PRs, I am noticing that the guidance given to it becomes evident. It does comment and document code much better than most human code I've seen and written, which is very welcome.
I see it's possible to reverse engineer the direction it was given and then match that with the change needed. So while the actual code and syntax become slightly less relevant, the structured and scoped direction of it starts taking protagonism, which was what we engineers always did before translating to syntax.