I assume we are talking about LLMs. Optimistically, it could help, given an appropriate, carefully crafted prompt. Realistically, most devs are lazy asses and as soon as AI gives them working code, they would move on to pumping out the next piece of crap that works. If we consider smarter AIs, that can actually understand code and logic, it could be much better. But here we should be more concerned about alignment of such AIs.