What makes you think so? We are talking about wrappers people can write around LLMs.

That has nothing to do with AIs in general. (Nor even with just using a single LLM.)