Comment by robertclaus

Conceptually this is wrapping an agent harness in an LLM call API. I wonder if this format is more digestible than the agent building tools the big labs are rolling out.