This discussion at VMware Explore 2026 examines securing agentic artificial intelligence across enterprise environments with a focus on identity governance and observability. Clayton Donley of Broadcom, vice president and general manager, Identity Management Security Division, appears on theCUBE Research with John Furrier of theCUBE and theCUBE video team to examine how enterprises adapt identity and security practices for agentic AI. Donley presents Broadcom's perspective on agent identities, API protection and distributed tracing, and they explain how existing application monitoring techniques and recent Broadcom capabilities such as AgentMinder apply to governing large fleets of AI agents in production. Donley emphasizes the imperative to assign identities to agents, centralize policy enforcement and apply observability to prompts and tool calls. They advocate starting with traffic monitoring, routing model access through enterprise keys and layering controls to prevent circumvention, which enables auditability, cost control and an improved security posture as agents scale across organizations. The conversation provides actionable guidance on identity management, API protection, distributed tracing and operational governance for agentic AI deployments.