# Google retools Kubernetes for AI agents with Agent Substrate instead of a new stack

published: 2026-11-13T04:05:38.303Z
topic: infrastructure
source: https://siliconangle.com/2026/11/12/agentic-workloads-push-google-retool-kubernetes-agents-kubeconna/
url: https://aiagentnews.io/stories/google-retools-kubernetes-for-ai-agents-with-agent-substrate-instead-of-a-new

Google Cloud says it is tuning Kubernetes clusters for agentic workloads rather than building a separate stack, according to executives speaking on theCUBE at KubeCon. Its Agent Substrate project runs AI agents in sandboxes directly on Kubernetes and has been donated to the Cloud Native Computing Foundation.

Google Cloud executives Jago Macleod and Jaana Dogan say agents strain Kubernetes in ways it was not designed for, including generating untrusted code, sitting idle for long periods and creating heavy pod churn that pressures the control plane. Macleod says Google's approach is to focus each cluster on the workload it runs and optimize it for that workload.

Google's answer for running agents is Agent Substrate, which runs AI agents in sandboxes on Kubernetes. Google donated the project to the Cloud Native Computing Foundation and recently released version 1.0. Dogan says agent code is untrusted by default and needs sandboxing and network guardrails.

Dogan also described a cost problem: customers' agents are often idle, yet a sandbox is allocated for the whole session, wasting CPU cycles. The segment was sponsored by Google Cloud, which has no editorial control over SiliconANGLE content.

## Why it matters if you build agents

If you run agents on Kubernetes, plan for sandboxed execution of agent-written code and for sessions that hold resources while idle.

Agent Substrate is now at version 1.0 under the CNCF, so teams wanting to avoid a custom control plane have an open-source option to evaluate.