Beyond Sandboxes: Architecting Durable Runtimes for AI Agents

This talk covers architectural patterns for building production-ready AI agents that persist state, checkpoint every operation, and recover from crashes without losing context. It introduces the late-bound saga pattern for agent orchestration and discusses how to implement retries, human-in-the-loop pausing, and auditability at scale. Designed for software architects and platform engineers building robust AI agent infrastructure.