> That is, an agent is a for loop which contains an LLM call. The LLM can execute commands and see their output without a human in the loop.

Am I missing something here?