Yeah, but this is self-correcting. Eventually it will get to a point where the data that you use to prompt the LLM will have more signal than the LLM output.

But if you get deep into an enterprise, you'll find there are so many irreducible complexities (as Stephen Wolfram might coin them), that you really need a fully agentically empowered worker — meaning a human — to make progress. AI is not there yet.