AI is increasingly part of DevOps work, even in teams that don’t think of themselves as “doing AI.” Models influence scaling, alerting, cost optimization, and developer workflows, often quietly and without clear ownership. In this talk, I’ll share real patterns I see when AI enters existing DevOps setups: responsibilities becoming blurred, “trust the model” replacing engineering judgment, costs and energy usage becoming harder to explain, and teams staying accountable for decisions they no longer fully control. This is a practical talk about what actually breaks when AI meets real-world DevOps —and how teams can adapt without adding more tools, more dashboards, or more process. The focus is on clarity, ownership, and keeping humans meaningfully involved.