Hard disagree. LLMs I simply next word predictors that's all they really do. It's the human driving them that makes the decisions.
LLMs don't "build" they simply do what they're told. So they're only as good as the engineer using them.
I think you're highly overrating them.