AI is starting to move beyond helping SREs investigate incidents and into actually taking action. That raises an important question: when is it safe to let an AI agent change something in production? This talk explores that transition from AI copilots to more autonomous SRE agents. We’ll look at practical ways to control that autonomy, including human approval, least privilege, policy enforcement, blast radius, dry runs, rollback, observability, and continuous evaluation. Using real-world incident scenarios, I’ll show where AI can add value today, where it should still stop and ask for help, and what needs to be in place before we let it act on its own. The goal is not to replace SREs. It is to reduce repetitive operational work while keeping production safe, observable, and under control.
Senior Site Reliability and Observability Engineer with extensive experience in enterprise infrastructure, cloud technologies, monitoring, automation, and reliability engineering. Specialized in observability platforms such as Instana, Dynatrace, Splunk among others, and in building scalable solutions that improve operational efficiency, visibility, and system resilience. Passionate about automation, infrastructure architecture, and the practical application of AI to solve complex technical challenges.
