As agentic systems become more autonomous, engineering teams must contend with unpredictable behaviour, cascading failures, unreliable tool interactions, changing model performance, and limited visibility into complex decision-making. In this panel, engineering leaders share the lessons learned from deploying AI agents in real-world environments. Explore the architectural patterns, evaluation strategies, and operational practices that help teams build agentic systems that remain reliable, observable, and resilient long after deployment. Key takeaways → The most common reasons agent systems fail in production and how to design around them. → How leading teams evaluate, monitor, and debug increasingly autonomous agents. → Architectural patterns that improve reliability, resilience, and observability at scale. → Lessons learned from deploying agentic systems in real-world production environments.