I meant "communicate with an AI *agent* to generate ...".

By agent I mean chaining the response of one LLM output into another, where each LLM has a specific "system" and set of tools to use for its specific job ... Though I don't know exactly what's going on under the hood of their source code.