Secure and Scalable AI on Ray + Kubernetes - Google's Decoupled Agent Pattern

Learn how to deploy agentic AI systems in production through Google's Decoupled Agent Pattern, a scalable architecture that combines Ray and Kubernetes for secure, resilient AI applications. Discover the core production challenges of agentic systems, including integrating LLMs, tools, and long-lived stateful agents while maintaining security, elasticity, and high-throughput execution. Explore how the Decoupled Agent Pattern addresses these challenges by separating stateful agent logic from stateless, scalable tools, with agent core logic running as durable Ray Actors managed by Ray's Global Control Store for high availability, while tools execute as thousands of stateless Ray Tasks for massive parallelism. Understand how untrusted or dynamically generated code runs in gVisor sandboxes for kernel-level isolation without compromising throughput through Kubernetes' secure runtime capabilities. Watch live demonstrations featuring a financial analysis agent running on a Ray cluster on Google Kubernetes Engine, and see how the architecture leverages KubeRay's topology-aware placement for optimal scheduling, intelligent capacity management with tools like Kueue for cost-efficient batch scheduling, and zero-downtime upgrades with fault-tolerant agent execution. Gain practical insights for building scalable, resilient, and secure agentic runtimes that combine Ray's distributed computing strengths with Kubernetes' security and orchestration capabilities.