- Joey Orlando When Agents Write the Code, Who Runs the Platform? - Sidhant Dixit Your CLI has a new user: the LLM - Margaret Gu Continuous Promotions: bridging the gap CI and CD in GitOps - Daniil Fedotov Self-Healing Infra as Code with AI Agents - Mohamed Ali Elserngawy
escape ≠ node compromise • Fine-Grained Kubelet Authorization — tighter node access • Gang Scheduling — atomic pod groups for ML/HPC • Metrics API graduates to stable (GA)
— autoscale all the way down In-Place Pod Vertical Scaling — resize without restart Alpha • • • • PodGroup API — decoupled gang scheduling Memory QoS — cgroup-based memory priority tiers DRA Enhancements — GPU health signaling & visibility Server-Side Sharded List/Watch — large cluster scalability
Metrics API graduates to stable (GA) • Pod-level resource requests/limits added • HPA scale-to-zero moves to Beta (on by default); nftables becoming default kube-proxy backend • DRA (Dynamic Resource Allocation) keeps expanding for AI/GPU workloads • Metrics API GA = stable base for kubectl top and HPA tooling • Pod-level resources cut wasted allocation across multi-container pods • Scale-to-zero + DRA show K8s adapting for cost-efficient AI/ML at scale
(announced Aug 5) • Cloud-native Global Server Load Balancer (GSLB) for K8s • Built on native K8s APIs, CoreDNS, ExternalDNS • Solves cross-cluster/cross-region availability and disaster recovery, no vendor lock-in • Proven in production — Millennium bcp reports 99.99% uptime • Incubation reflects maturing governance, security, and adoption