That matches my experience with LLM-aided PRs - if you see a useEffect() with an obvious LLM line-comment above it, it's 95% going to be either unneccessary or buggy (e.g. too-broad dependencies which cause lots of unwanted recomputes).