Google releases Agent Substrate 1.0 to sandbox AI agents on Kubernetes
Google donates Agent Substrate to CNCF to handle agentic workload strain without fragmenting the open-source Kubernetes ecosystem.
Key points
- Agent Substrate version 1.0 is now available and has been donated to the Cloud Native Computing Foundation.
- The project runs AI agents in sandboxes directly on Kubernetes to manage untrusted code generation.
- Google engineers state agentic workloads strain the control plane through high pod churn and idle periods.
Google Cloud has donated Agent Substrate to the Cloud Native Computing Foundation and released version 1.0, according to the published report The project enables AI agents to run inside sandboxes directly on Kubernetes clusters.reported This move addresses the specific operational strain that agentic workloads place on existing infrastructure.Google has released a tool that lets AI agents run safely inside standard Kubernetes clusters.Google donated Agent Substrate 1.0 to the CNCF to sandbox agentic workloads on Kubernetes.
Evidence: reported
Google retools Kubernetes for AI agents instead of building a new stackaifleet-siliconangle
How we got here
Kubernetes was originally designed for microservices, then adapted for training jobs where single pods occupy entire nodes, Jago Macleod explained. Agentic applications behave differently, generating untrusted code dynamically and sitting idle for extended periods. This pattern creates a level of thrash and churn that differs significantly from the orchestrator’s original design intent.
What is new
Google is tuning clusters to be optimized for specific workloads rather than treating them as infinitely flexible engines, Macleod stated. Agent Substrate provides the necessary sandboxing and network guardrails for code that agents write themselves. This approach allows platform teams to maintain their existing Kubernetes compute fabric without building custom control planes.
What to watch
Cloud providers are rethinking how general-purpose a cluster should be in response to these new workload patterns, according to the report.reported The focus is shifting toward efficiency due to constraints on memory and machine availability. Google aims to prevent fragmentation of the open-source ecosystem while accommodating these distinct operational requirements.
Evidence: reported
Google retools Kubernetes for AI agents instead of building a new stackaifleet-siliconangle
Availability and limits
Agent Substrate version 1.0 is available through the Cloud Native Computing Foundation, as according to the published report It runs directly on Kubernetes and provides sandboxing for untrusted code generated by agents.reported The solution is designed to handle high pod churn and idle periods without requiring a separate control plane. No specific pricing or regional availability details were provided in the source material.
Evidence: reported
Google retools Kubernetes for AI agents instead of building a new stackaifleet-siliconangle
Why it matters
- Developers can run agentic workloads on existing Kubernetes clusters without fragmenting their infrastructure.
- Sandboxing addresses the security risk of agents generating untrusted code on the fly.
- The solution mitigates control plane strain caused by rapid pod churn and idle periods.
- Donation to CNCF suggests a path toward standardized support for agentic workloads.
Topics: Kubernetes, AI agents, Google Cloud, CNCF, infrastructure
Updated
Then / now / next
Teams built custom control planes to cope with agentic workload strain on Kubernetes.
Google released Agent Substrate 1.0 to sandbox agents directly on Kubernetes.
Cloud providers will continue rethinking cluster general-purpose design for efficiency.
Story timeline
- Then Teams built custom control planes to cope with agentic workload strain on Kubernetes.
- 12 Nov 2026 Google retools Kubernetes for AI agents instead of building a new stack
- Now Google released Agent Substrate 1.0 to sandbox agents directly on Kubernetes.
- Next Cloud providers will continue rethinking cluster general-purpose design for efficiency.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
Agent Substrate
A CNCF project that runs AI agents in sandboxes on Kubernetes.
Control plane
The component of Kubernetes that manages cluster state and scheduling.
Pod
The smallest deployable unit in Kubernetes, often containing one or more containers.
Questions readers ask
What is Agent Substrate?
It is a project donated to the CNCF that runs AI agents in sandboxes on Kubernetes, currently at version 1.0.
Why do agents strain Kubernetes?
Agents generate untrusted code, sit idle for long stretches, and churn through pods at a pace that strains the control plane.
Is this a new stack?
No, Google is retuning existing Kubernetes clusters for specific workloads rather than building a new stack.
Connected stories
Reading receipt
- Agent Substrate version 1.0 is now available and has been donated to the Cloud Native Computing Foundation.
- The project runs AI agents in sandboxes directly on Kubernetes to manage untrusted code generation.
- Google engineers state agentic workloads strain the control plane through high pod churn and idle periods.
Still open: How does Agent Substrate impact overall cluster performance metrics?
Still open: Are there specific configuration guides for integrating with existing GKE clusters?