Google has announced that the GKE Agent Sandbox is now generally available, marking a significant milestone for developers working with AI agents. This cloud-native execution environment is designed to provide a secure and scalable foundation for deploying agent workloads.
Since its preview at KubeCon NA in November 2025, GKE Agent Sandbox has seen rapid adoption, with a remarkable 16x growth in usage on Google Kubernetes Engine (GKE) within just five months. This growth is supported by collaborations with key customers such as Langchain and Lovable, who are utilizing the platform to deploy millions of agents in production.
Key Features of GKE Agent Sandbox
The GKE Agent Sandbox offers several essential features tailored for modern agent workloads:
- Pod Snapshots: This feature allows agents to suspend idle workloads and resume them quickly, optimizing compute resource usage.
- Low Latency Provisioning: The integrated warm pool enables rapid allocation of sandboxes, achieving sub-second latency for most requests.
- Cost-Effective Warm Pool: By utilizing standby capacity buffers, the warm pool can be maintained at a lower cost while ensuring quick replenishment.
- Enhanced Security: The sandbox supports gVisor and customizable kernel isolation, ensuring robust security and isolation for workloads.
In addition to the general availability of GKE Agent Sandbox, Google introduced Agent Substrate, an open-source project aimed at enhancing the performance and density of agent workloads. This new layer of abstraction allows for real-time movement of agents onto and off of available compute resources, optimizing Kubernetes capabilities for ultra-scale demands.
Why Agent Substrate Matters
Agent Substrate is designed to handle the complexities of scaling agent workloads, which can reach tens or hundreds of millions of instances. By addressing the limitations of the Kubernetes control plane, it enables faster execution and better resource management for high-frequency tool calls.
As the demand for compute resources continues to grow, both GKE Agent Sandbox and Agent Substrate are positioned to provide the necessary infrastructure for developers to build and manage autonomous agents effectively.
Getting Started
Developers are encouraged to explore the capabilities of GKE Agent Sandbox for their workloads and to participate in the open-source community for Agent Substrate. This collaborative effort aims to shape the future of agent infrastructure, ensuring that it meets the evolving needs of the industry.
- Try GKE Agent Sandbox for your agent workloads.
- Join the open-source community to contribute to Agent Substrate.