By the same logic, a hallucinating LLM is also overkill versus just doing the simple task yourself and not needlessly adding risk to it.

The point still remains: let's see what the LLM delivered that the user actually used.