Proposal¶

The submission to Cloud Native Days Norway 2026. Source (login required).

Title¶

Agentic Cloud Ops: Praktisk AI-drevet Kubernetes-drift (Agentic Cloud Ops: Hands-on AI-driven Kubernetes operations)

Session details¶

Field	Value
Format	Workshop
Language	Norwegian
Level	Intermediate
Duration	2 hours
Audience	Developers, architects, DevOps engineers, managers
Topics	Kubernetes & Orchestration · Security & Governance
Speakers	Lars Effenberger (leffen), Lars Søraas

Abstract¶

As submitted (Norwegian — unedited)

Er du klar for å gi slipp på kontrollen og la AI-agenter drifte clusteret ditt? I takt med at AI-modeller blir smartere, går vi fra "Infrastructure as Code" til "Infrastructure as Agents". I denne super-praktiske, to-timers workshopen dykker vi ned i "Agentic Cloud Ops". Vi skal se på hvordan vi kan bruke agenter til å ikke bare deploye applikasjoner, men til å overvåke, feilsøke og sikre (hardne) Kubernetes-miljøer helt autonomt.

Vi starter med å spinne opp et lokalt cluster i Docker (Talos/k3s), før vi slipper agentene løs. Du vil lære hvordan du setter opp agentene, bygger tillit til dem, og hvilke reelle farer og muligheter som ligger i denne teknologien. Hva skal egentlig til for å lykkes med agentisk bruk i produksjon? Bli med, så finner vi det ut i praksis!

Translation. Are you ready to let go of control and let AI agents operate your cluster? As AI models get smarter, we're moving from "Infrastructure as Code" to "Infrastructure as Agents". In this deeply practical, two-hour workshop we dive into "Agentic Cloud Ops". We look at how to use agents not just to deploy applications, but to monitor, debug and harden Kubernetes environments fully autonomously.

We start by spinning up a local cluster in Docker (Talos/k3s), then turn the agents loose. You'll learn how to set the agents up, how to build trust in them, and what the real dangers and opportunities of this technology are. What does it actually take to succeed with agentic operations in production? Join us and let's find out in practice.

Takeaways (as submitted)¶

An understanding of Agentic Cloud Ops vs. traditional GitOps/IaC.
Hands-on experience using agents for deployment, debugging and security hardening in K8s.
Insight into risk management and success criteria for autonomous systems.

Session outline (as submitted)¶

Note from the submission: tested with Claude (very good results), Codex works acceptably, Ollama untested at this time.

Intro: What is Agentic Cloud Ops? (15 min) — the concept (action → observation → decision), the opportunities (faster MTTR, automatic optimisation), the dangers (hallucinations deleting namespaces, lack of determinism, runaway loops), and the success criteria (human-in-the-loop, good RBAC, strong system prompts).
Bootstrap: local cluster (15 min) — make cluster-up, verify nodes and context, walk through the repo structure.
Lab 1: The first agent (30 min) — have an agent deploy Nginx with 3 replicas behind a NodePort; it must verify pods itself and debug a deliberately planted ImagePullBackOff.
Lab 2: Security and hardening (35 min) — the cluster is deliberately insecure. Put the agent in "Hardening Mode"; it adds securityContext, resource limits and tightened pod permissions without breaking the application.
Lab 3: Chaos and recovery (15 min) — make chaos breaks something; the agent must find it, explain it and fix it, with no manual kubectl from participants.
Q&A and wrap-up (10 min) — how did it go, what scared us, how to take this to production (gradual adoption, read-only vs. read-write agents), make cluster-down.

The outline has been revised

The submitted outline has zero timing slack and some unresolved choices. See the Premise review for the analysis and the revised schedule for what we actually intend to run.

Status¶

Proposal submitted
Co-speaker invited and accepted (Lars Søraas)
Slides uploaded
Repo and lab content finished