LLM vs Agent and the Agent Loop