I don't either.

But what do you believe about agent harnesses? About AIs that have tools that they can call that can do tasks that the AI, by itself, is not good at? Is that enough of "not LLM alone"? I don't know.