AI agents don’t just execute tasks and exit. Once deployed, they run continuously, accumulate state, interact with changing systems, and influence future behavior in ways that aren’t always visible at launch. This session examines what it’s actually like to operate AI agents in production. We’ll discuss how long-running agents introduce new operational challenges around state growth, behavioral drift, upgrades, and incident response. The focus is on the day-to-day realities engineering teams face once agents are live, and what changes when systems must be managed over weeks and months rather than individual executions. Key takeaways: →How long-running AI agents accumulate state and why this makes behavior harder to reason about over time. → The operational challenges that emerge once agents are live, including drift, upgrades, and incident response. → What engineering teams must plan for when managing agents as ongoing systems rather than one-off executions.