kagent tutorial: debug Kubernetes services and autoscaling

This hands-on tutorial compares manual Kubernetes troubleshooting with kagent on the same two cluster failures. The presenter describes kagent as an open source AI agent that runs inside a Kubernetes cluster, with YAML configuration and scoped access to tools. The demonstration does not establish that its language model runs locally or works offline.

The first exercise follows an order-api service that cannot reach its running pods. Checking endpoints reveals that the service has found none: its selector expects version V1, while the pods carry V2 labels. Changing the selector to V2 restores endpoints and a successful curl response. In the kagent chat, a diagnostic prompt identifies the mismatch, and a second prompt requests the change.

The inventory exercise concerns an HPA configured for three to ten replicas, with only one pod running. Manual checks find working node metrics, then namespace events expose a pod quota capped at one. The presenter raises the quota to ten and restarts the deployment. Given the same symptom, kagent returns evidence and ranked possible causes, including the quota, then asks how to proceed. This portion shows diagnosis rather than a completed agent repair.

The presenter reports about two minutes per agent investigation. Those timings describe this lab, rather than a general performance guarantee.