Human review is even more important for LLM-generated code, because LLMs are so much worse at writing code than humans are. You cannot under any circumstances just let them rip or your results will suck.