What if your infrastructure could heal itself before anyone noticed it was broken? I want to show you a custom-built AI-native operations agent that monitors services and infrastructure and fixes common incidents before users notice anything is wrong. We will explain how we built it using Microsoft Teams as the control surface, large language model (any really) as the reasoning engine, and Model Context Protocol (MCP) to connect the agent to Docker, Azure and VMware vCenter. Tiered autonomy lets agent fix low-risk issues on its own while asking for human approval on everything else. In a live demo, we will break production on stage and watch agent detect, diagnose and restore it from a single phone conversation.