1 and 2 are really only an issue if you vibe code. There's no reason to expect properly reviewed AI assisted code to be any worse than human written code. In fact, in my experience, using LLMs to do a code review is a great asset - of used in addition to human review