Nishant Gupta talks production evals for agentic AI at Meta

Traditional offline benchmarks and static datasets fail for autonomous agents. The talk covers approaches to handle complexity, non-determinism, and operational risks in real-world evaluations.