[ryancoleman]

That's basically how it works! (with human authored functions that validate the result, automatically providing feedback to the LLM if needed)