Build phase 4Planned
kagent Integration
Migrate the custom Anthropic-SDK agent to kagent, a CNCF Sandbox framework, so the assistant becomes a Kubernetes custom resource instead of a bespoke Deployment.
Rolling your own agent proves you understand the mechanics. Migrating it to a CNCF-native framework proves you know the ecosystem and when to stop maintaining bespoke plumbing. Showing that contrast and documenting the trade-off is the point of this phase.
Deliverables
- kagent operator installed on the cluster via Argo CD.
- Agent as a CRD: the K8s assistant reimplemented as a kagent
Agentcustom resource, not a custom Deployment. - MCP tools: at least two registered, a Prometheus PromQL query tool and a kubectl-style read tool.
- Architecture Decision Record: “Why kagent over the raw SDK,” documenting what it gives and what it costs.
Why migrate at all
The raw SDK agent works. kagent replaces the hand-written Deployment, config, and tool wiring with a declarative
Agent resource that Argo CD manages like everything else. It connects to tools over MCP, and the ADR captures that reasoning honestly, including the trade-offs.Tasks
KAGENT-001KAGENT-002KAGENT-003KAGENT-004KAGENT-005KAGENT-006KAGENT-007KAGENT-008KAGENT-009
Write-ups
- “kagent vs rolling your own: migrating my AI agent to a CNCF framework.”
- “MCP tools on Kubernetes: connecting AI agents to Prometheus and kubectl.”