In most software systems, the testing process ends at deployment. But what if testing didn’t stop — and instead evolved in real-time, guided by what users actually do in production? This talk explores how AI agents can monitor logs, telemetry, and user interactions to detect anomalies, learn system behavior, and dynamically generate new test cases. We’ll dive into real-world techniques for integrating generative models into observability stacks, closing the loop between production and pre-prod environments. Along the way, we’ll examine tools, patterns, and risks of letting AI take part in live QA—turning your test strategy into a living, adapting system.