Add to that the million inaccuracies... manually reviewing LLM results instead of blindly believing them takes about as much time as not using the LLM in the first place. But most engineers are in the blind believer camp, as such, the architecture & code become rubbish, the documentation unreadable.
