Modern observability stacks based on OpenTelemetry and Grafana provide powerful signals across metrics, logs, traces, dashboards, and alerts. However, engineering teams still spend a significant amount of time correlating these signals with runbooks, internal documentation, Kubernetes context, incident history, and operational tools before they can decide what to do next.

This session explores how AI agents and the Model Context Protocol can extend observability workflows from passive monitoring to guided operational action. Starting from a real enterprise scenario, we will discuss how agents can safely consume observability data, reason over production context, and interact with tools through governed interfaces.

The talk will cover architectural patterns, MCP-based tool access, Grafana and OpenTelemetry integration points, Kubernetes context, security boundaries, human-in-the-loop design, and practical lessons from bringing agentic workflows closer to production operations.

Attendees will leave with a practical view of how AI can augment observability without replacing engineering judgment.