LLMs produce statistically plausible text given a particular context.

«Agents» run that in a loop, «given everything we’ve said so far and this new text, what is the most plausible next piece of text?»

There is nothing there.
It’s all in people’s heads.