KubeAutoDay US 2026 - Jev-ops

Transcript

1. Kubernetes, Meet Jev: Fast Decisions, Real Pods Pablo Fredrikson Principal Engineer
2. Avengers (2012)
3. Avengers (2012)
4. Reasoning models and System One • Reasoning models ◦ Explore a problem and develop a multi-step solution. • System One / Jev ◦ Evaluate focused questions. Return typed decisions.
5. What is Jev? • • • • TypeSafe’s model for fast, structured decisions Input: state + questions with allowed answers Output: typed values with uncertainty signals And it does it extremely fast and extremely cheap
6. Live demo: a Kubernetes cluster that needs to scale • Nginx pods receiving normal traffic • We are going to try to scale up and down
7. Demo
8. Live demo 2: a Kubernetes cluster that needs to heal • HTTP 500s: healthy pods, failing application • CrashLoop: detect, restart, verify • Watch the decision, confidence and measured latency
9. But before demo #2
10. But before demo #2 hi 👋
11. Pablo Fredrikson • • • • Principal Engineer @ Bitso 20+ years of experience in the field Professional Nerd Pelado Nerd @ YouTube
12. Demo #2
13. How the demo works 1. Node collects logs, pod state and events via kubectl. 2. “Analyze with Jev”: a. kubectl get pods -A | jev-ops analyze kubernetes b. jev-ops analyze demo-app --pack-dir demo/live-cluster/packs --provider typesafe --json 3. Typed JSON returns to Node and the dashboard. 4. Node policy restarts pods or escalates to Slack.
14. The future of LLMs • System One and reasoning models might merge or at least work closely together. ◦ The harness or model can route each request to different models, transparent to the user.
15. Thanks!