Catastrophic Agent Failure and How to Avoid It

Learn to identify, prevent, and mitigate catastrophic failures in AI agent systems through this 25-minute conference talk from Agents in Production 2025. Discover practical strategies for building resilient automation systems as Edward Upton shares insights from his experience developing production-ready browser agents and combating online fraud at scale. Explore real-world examples of high-stakes agent failures across industries including fraud detection, healthcare, and insurance, while understanding the critical infrastructure and evaluation frameworks needed to ensure reliable performance. Gain expertise in creating trustworthy autonomous systems that can handle enterprise-grade applications without compromising security or dependability, drawing from lessons learned in both agent development and fraud prevention for major banks and brands.