1. create a goroutine 2. call the llm api in a loop 3. if you get back a tool invocation then do the tool 4. if you get back the answer send it to the caller using a channel since this style is so good for ai agent actions we could give it a name, like “actor model”