Google Announces Agent Compute Engine for Secure AI Agent Execution

InfoQ (AI, ML & Data)
Read full story· 3 min
Google launched Agent Compute Engine, a managed service running AI agents in isolated microVMs with secure credential handling and fast checkpointing to suspend and resume environments. It supports diverse workloads including interactive agents, autonomous software engineering, and large-scale model evaluation, and is used internally for Gemini and Antigravity platforms.

What happened

Google has introduced Agent Compute Engine, a managed service designed to run AI agents in isolated microVMs. The service separates sensitive credentials from agent execution environments and uses an external gateway to manage authentication and secret injection. It features fast checkpointing to pause and resume agent environments efficiently. Google currently uses this platform internally for its Gemini and Antigravity products, and plans to offer a preview to customers in the coming months.

Key details

  • Agent Compute Engine runs AI agents in microVMs with isolated sandboxes.
  • Credentials and API keys are kept outside the agent environment and managed via Agent Gateway.
  • The service supports interactive agents, autonomous software engineering, model evaluation, and dynamic logic for Google services.
  • Google uses copy-on-write memory sharing and fast checkpointing to minimize resource use and latency.
  • Agent Compute Engine is not yet available to the public and pricing has not been disclosed.

Why it matters

Organizations using Google Cloud will soon have access to a managed service for running AI agents with enhanced security and control over credentials. This allows businesses to deploy AI workloads with reduced risk of credential exposure and more efficient resource management. Developers will be able to manage environment images, identities, and network access, while Google handles the underlying infrastructure and governance.


More in Agents