Keep in mind that our current theories in Neuroscience broadly agrees something similar is going on with mammalian, even reptilian brains. Hell, maybe even worm brains.

There's autonomous systems everywhere that calls each other for updates and in some certain brains, enough complexity that something that can called thinking occurs.

Practically, offloading calculations to a python REPL, machine translation to GTranslate API call, and knowledge search to Wikipedia corpus is going to let LLMs do what they do best - mask users intent and generate believable enough corpus. Let the facts stay factual and the hallucination stay hallucination.