I think a baseline requirement would be that it… thinks. That’s not a thing LLMs do.

That’s an odd claim, given that we have so-called thinking models. Is there a specific way you have in mind in which LLMs are not thinking processes?