Agentic DevOps with Claude Code

4

Building the Platform Observability Plane

An AI platform without telemetry is a platform that can produce surprising answers and very little evidence. The observability problem starts before the first model request. We need reliable metrics, logs, and traces for Kubernetes, controllers, gateways, portals, and ordinary workloads. Only then does it make sense to add agent and model attributes to the same pipeline. Otherwise, the first AI incident becomes a scavenger hunt across five consoles and one very confident summary.

In this chapter, we will design a shared OpenTelemetry pipeline that separates telemetry producers from storage backends. We will configure collection for managed EKS, where the control plane is not scraped like a self-managed ...