While many organizations have proven what's possible with agentic AI, far fewer have solved the realities of deploying autonomous systems that can be trusted at scale. This panel brings together engineering leaders who have moved beyond prototypes to discuss the technical and operational challenges that emerge in production. From orchestration failures and unreliable tool use to observability, evaluation, governance, and cost control, we'll explore why agent systems break in the real world and the engineering practices that keep them running.