The problem for me is to predict what AI might get wrong. Claude can solve hard coding problems one day just to fail with basic stuff like thread safety the next. But overall I think it is clear that we reached the point where AI, if used correctly, saves a lot of development time.