No no, let's just stop thinking entirely and paste conversations from LLMs back and forth to each other. Then we'll use an LLM to summarize the conversation to tell us what was said. Then we'll use an LLM to do what was said. Then we can ask an LLM if what was done works.