Kubernetes v1.36 (released April 22) extends the kubelet's Topology, CPU, and Memory Managers to operate at the pod level rather than the container level, enabling more precise resource allocation for latency-sensitive AI/ML workloads. User Namespaces, which isolate workload UIDs from host UIDs and substantially narrow container-escape attack surface, reached general availability in this release. A companion post published April 30 announced that In-Place Vertical Scaling for Pod-Level Resources graduated to beta, allowing CPU and memory limits to be adjusted without restarting a pod. For platform teams running inference or training jobs the Pod-Level Resource Managers alpha is immediately worth evaluating: co-located pods competing for NUMA-aligned memory or dedicated CPU cores will benefit from the coordinated scheduling decisions the new managers enable. The User Namespaces GA removes the last major blocker for enabling the feature in hardened production clusters, and the vertical-scaling beta reduces disruption overhead for right-sizing long-running workloads. No immediate migration action is required, but clusters already on v1.36 should review the feature gates to opt in where appropriate.