At Metropolis, AI systems operate across physical locations, live decision points, and high-volume customer interactions, supporting more than 1M new members each month and saving over 212M minutes. At that scale, reliability is not just about model performance - it is about how the full system responds when conditions keep changing. This session explores what happens when agentic systems move from designed workflows into real operations: where behaviour starts to shift, why edge cases compound at scale, and what teams can do to keep systems stable when failure has immediate impact. Key takeaways: → What changes when agentic systems operate in real-world, high-volume environments → Where workflows become harder to predict, debug, and control at scale → How teams can build for reliability when conditions are constantly changing