Whenever you're ready, walk me through how you'd approach this.
What metrics do you prioritize when evaluating the performance of an autonomous agent in a production environment?
Discuss how you would select metrics for task success, response quality, safety, reliability, latency, and cost. Explain how you would combine offline evaluation, production monitoring, human review, and user feedback, while accounting for non-deterministic behavior and changing workloads.