The llm-d project introduces co-operative time-slicing to interleave independent reinforcement learning jobs on shared hardware, increasing aggregate accelerator duty cycles from ~40% to 70% without i…
The rise of AI agents is forcing a rethink of cloud infrastructure, as Kubernetes' design for short-lived, stateless requests fails to support the long-running, stateful reasoning loops of autonomous …
Kubernetes and Amazon EKS are ill-suited for AI agent workloads because the container orchestration model was designed for stateless HTTP requests, not long-running, stateful agent processes, accordin…
General availability of GKE Agent Sandbox, an open-source, Kubernetes-based execution environment designed for secure and scalable AI agent workloads, featuring capabilities like pod snapshots for idl…