Skip to content
Stack Courier SC markSCSTACKCOURIER
Google tunes Kubernetes for AI agents via Agent Substrate
Depth
1 outlet · newest source 87h ago
Agent builders

Google tunes Kubernetes for AI agents via Agent Substrate

Google Cloud engineers explain how they are optimizing Kubernetes clusters for agentic workloads to avoid control plane strain and fragmentation.

Key points

  • Google released Agent Substrate version 1.0 to run AI agents in sandboxes on Kubernetes.
  • The project was donated to the Cloud Native Computing Foundation by Google.
  • Agentic workloads create high pod churn that strains the Kubernetes control plane.

Google Cloud is retooling Kubernetes for AI agents rather than building a new stack, according to a report by SiliconANGLE AI. Jago Macleod, director of engineering for Kubernetes and Google Kubernetes Engine, stated that the company is tuning clusters to be optimized for specific workloads instead of treating them as infinitely flexible engines.Google Cloud engineers are adjusting Kubernetes settings to better handle the unique demands of AI agents.Google Cloud is implementing workload-specific cluster tuning via Agent Substrate to mitigate control plane API limits caused by agentic pod churn.

Evidence: reportedGoogle retools Kubernetes for AI agents instead of building a new stack

aifleet-siliconangle

What was released or claimed

Google donated the Agent Substrate project to the Cloud Native Computing Foundation and released version 1.0. Jaana Dogan, principal engineer at Google Cloud, explained that this runtime allows AI agents to operate in sandboxes directly on Kubernetes. Because agents generate untrusted code on the fly, Dogan noted that the code requires sandboxing and network guardrails by default.

Evidence: reportedGoogle retools Kubernetes for AI agents instead of building a new stack

aifleet-siliconangle

Who it affects

Platform teams that have spent a decade making Kubernetes their compute fabric are affected. Dogan reported that customers told Google they did not want agentic workloads to fragment the open-source ecosystem. The changes also impact developers who previously resorted to building custom control planes to cope with the thrash and churn of agent pods.

Evidence: reportedGoogle retools Kubernetes for AI agents instead of building a new stack

aifleet-siliconangle

What is still unknown

It is unclear how many organizations have already adopted Agent Substrate version 1.0. The material does not specify which other cloud providers are implementing similar optimizations for their Kubernetes offerings. It is also unknown whether the sandboxing approach introduces measurable latency for agentic workloads compared to naive implementations.

Evidence: reportedGoogle retools Kubernetes for AI agents instead of building a new stack

aifleet-siliconangle

The takeaway

Macleod stated that there is insufficient memory and machine capacity to support general-purpose clusters for all workloads. He argued that focusing the Kubernetes cluster on the specific workload it is running increases efficiency. This approach aims to prevent the API machinery from running into limits caused by excessive pod churn.

What changes for users

Platform teams can use Agent Substrate version 1.0 to run AI agents in sandboxes on Kubernetes. Users gain access to network guardrails for untrusted code generated by agents. Availability is tied to the Cloud Native Computing Foundation project, which Google donated. Limitations include the need for specific cluster tuning rather than relying on general-purpose configurations.

Evidence: reportedGoogle retools Kubernetes for AI agents instead of building a new stack

aifleet-siliconangle

Why it matters

  • Teams can avoid building custom control planes to handle agentic workload churn, saving engineering resources.
  • Sandboxing untrusted agent code directly on Kubernetes reduces the need for separate infrastructure layers.
  • Optimizing clusters for specific workloads addresses memory and machine constraints cited by Google engineers.
  • Donating to CNCF suggests a path to standardize agentic runtime support without fragmenting the ecosystem.

Topics: Kubernetes, AI agents, Google Cloud, CNCF, infrastructure

Updated

For readers

Developers

  • Agent Substrate version 1.0 allows running agents in sandboxes on Kubernetes.
  • Untrusted code generated by agents requires network guardrails by default.

Business

  • Google is optimizing clusters for specific workloads to improve efficiency.
  • Customers preferred not to fragment the open-source Kubernetes ecosystem.

Researchers

  • Agentic workloads create thrash and churn that Kubernetes was not built for.
  • Custom control planes have been used by some teams to cope with agent pods.

Ask the article

Local keyword search of this article; no generated answers.

Did this article give you the evidence you needed?

No vote saved on this device.
Your choice stays on this device. No public tally.

Questions readers ask

Why is Kubernetes struggling with AI agents?

Agents generate untrusted code on the fly and churn through pods at a pace that strains the orchestrator's control plane, according to the published report

What is Agent Substrate?

It is a project released by Google that runs AI agents in sandboxes directly on Kubernetes, donated to the Cloud Native Computing Foundation.

Did Google build a new stack for agents?

No, Google retooled Kubernetes instead of building a new stack, focusing on optimizing clusters for specific agentic workloads.

Connected stories

This storyGoogle opens agent tools to help Kubernetes manage idle AI workloadsGoogle opens agent tools to helpGoogle Cloud's Project Chrysalis slashes outage fixes to 1.8 seconds with governed AI agentsGoogle Cloud's Project ChrysalisNYT Recommends AI Agents for Job Hunting and Restaurant ReservationsNYT Recommends AI Agents for JobGoogle ships A2UI 1.0 protocol for native generative UI agentsGoogle ships A2UI 1.0 protocol f

Reading receipt

  • Google released Agent Substrate version 1.0 to run AI agents in sandboxes on Kubernetes.
  • The project was donated to the Cloud Native Computing Foundation by Google.
  • Agentic workloads create high pod churn that strains the Kubernetes control plane.

Still open: How does the performance of sandboxed agents compare to naive implementations with many pods?

Still open: Will other cloud providers adopt similar cluster-tuning strategies for agentic workloads?

Time spent: 0:00 · Depth: Full

Sources

  1. SiliconANGLE AI

Related stories