I know you haven't heard from me in a minute, but I've been making you some useful content, and here's just a taste of what I've been up to:
1. I was in San Francisco for just 30 hours to interview experts on the Agentic DevOps podcast at the AI Security Summit, and it was awesome and impactful. I'll share more when those episodes drop soon.
2. I've got a backlog of podcast episodes and videos to share, so expect more frequent newsletters as I catch up with the backlog.
3. Will you be at KubeCon NA in Salt Lake City next month? I'll be speaking on Automation and Kubernetes at KubeAuto day, come say hi 👋. It's a free pre-conf event.
4. More KubeCon: I'll be around the vendor booth area all week; let's get a selfie! 🤳 I'll be recording podcast episodes at the Cast AI booth, and I got to make a custom tabletop game, "AI Apocalypse Defenders" that we've made 100 copies of to give away at the Chainguard booth during KubeCrawl.
Live this week: Fix your resource over-provisioning
Cast AI Staff Solutions Architect Steve Fraser joins me Thursday for an hour of learning and implementing ways to automate node and pod resource optimization. Sign up, watch live or the recording, and enter the raffle for AirPods Pro 3 and Cast AI swag.
Why Attend?
Average CPU utilization across Kubernetes clusters is around 8%, and 69% of clusters request more CPU than their workloads use. If you run Kubernetes, some of this probably sounds familiar:
- Resource requests still carry the headroom someone added at launch
- The node autoscaler scales to requests, so padded requests mean extra nodes
- Half-empty nodes stay up because the pods on them are stateful or long-running
- Spot stays off after one interruption took something down
- Automation stays off production until someone proves it’s safe
We’ll walk through how to fix each of these and what the rollout looks like, starting in read-only mode and enabling automation cluster by cluster.
Source: Cast AI 2026 State of Kubernetes Optimization report
What You’ll Learn
- How to set CPU and memory requests from real usage and resize pods in place, without restarts
- How the node autoscaler picks cheaper instances, packs pods tighter, and handles Spot
- How live migration moves stateful and long-running pods with minimal disruption, so underused nodes can be consolidated
- How to add these optimizations on top of Karpenter or Cluster Autoscaler without changing how nodes get provisioned
- What savings and utilization gains teams see once automation is running
Thanks for reading, and you'll be hearing from me again soon,
Bret
