Build phase 3Planned

AI Agent (Anthropic SDK)

The phase where the portfolio stops being 'yet another K8s cluster' and becomes the K8s + AI intersection story: a Kubernetes-native assistant running on the platform.

A new repo, k8s-ai-agent, hosts a Python agent built on the Anthropic SDK using the tool-use pattern. It runs as a pod on the cluster, deployed by Argo CD, and can answer natural-language questions about cluster state (pod status, logs, events) and help troubleshoot.

The demo scenario

Inject latency or errors into crashdummy, then ask the agent what's wrong. It reads the telemetry and cluster state and explains which workload is failing and why. That's the whole thesis in one interaction.

Deliverables

Agent v0.1.0: a K8s assistant that reads pod logs, describes resources, and troubleshoots via natural language.
Least-privilege RBAC: a dedicated ServiceAccount with a minimum-privilege ClusterRole. Never cluster-admin.
GitOps deployment: Deployment + Service managed by Argo CD, config via ConfigMap, API keys via External Secrets.
Health & metrics: liveness/readiness probes, resource limits, and a /metrics endpoint scraped by Prometheus.
Grafana dashboard: agent request count, latency, and error rate.

Tasks

AGENT-001AGENT-002AGENT-003AGENT-004AGENT-005AGENT-006AGENT-007AGENT-008AGENT-009AGENT-010AGENT-011AGENT-012AGENT-013

Write-ups

“My first AI agent running in Kubernetes: architecture and lessons learned.”
“RBAC for AI agents on Kubernetes: why your agent shouldn't be cluster-admin.”