As organizations deploy increasingly complex agentic systems, evaluating performance has become one of the biggest barriers to production readiness. Unlike traditional software, agent behaviour can vary across tasks, environments, and interactions, making quality difficult to measure and maintain. This panel examines how leading teams are approaching evaluation at scale. From automated assessments and benchmark design to human review and production monitoring, we'll explore the frameworks, metrics, and processes being used to understand agent performance and identify regressions before they impact users. The conversation will address the growing challenge of establishing trust in agentic systems through rigorous, repeatable, and scalable evaluation practices.