In this KubeCon + CloudNativeCon North America 2025 segment, Google Cloud’s Brandon Royal and Mofi Rahman join analyst Rob Strechay and theCUBE’s Savannah Peterson as part of a special series celebrating 10 years of Google Kubernetes Engine (GKE). The group cuts through the agentic AI buzz by grounding the definition of agents – LLMs with tool and function access – in the realities of Kubernetes, security and infrastructure complexity, highlighting why this community cares about what’s real, not vaporware. Royal and Rahman dive into Google’s investments in isolation with technologies such as gVisor and introduce Agent Sandbox on Kubernetes, giving teams a secure environment for code execution and “computer use” so agents can browse, test and experiment safely. They explain how sandboxed browser automation, pod snapshotting and fast restore deliver roughly 90% faster cold-start performance, enabling thousands of agents to spin up in sub-second time, while a Python SDK lets AI engineers plug in from familiar frameworks without being Kubernetes experts. Strechay adds why templated, operator-style patterns and reduced toil for platform teams are key to making agentic workflows repeatable, secure and ready for next year’s real-world customer stories.