LLMs are good for documenting specific things.

E.g., "find where the method X is called and what arguments are passed".

That can be useful for refactoring or debugging.

Coding is the worst way to use an LLM though.