AboutAI ToolsCursorClaude CodeLovableBolt.newAI tool reviewsTutorialsVibe codingAI agentsMobile apps with AIAll tutorialsBlogSponsorWork With MeJoin 285K+ Subscribers

6 November 2026 · 4 min read

Google Cloud launches Agent Compute Engine for isolated AI agents

Google Cloud has introduced Agent Compute Engine, a microVM service for running AI agents with hardware isolation and idle optimisations, alongside open-source tools and governance controls.

On 6 November 2026, Google Cloud introduced Agent Compute Engine, a compute service designed to run AI agents in isolated micro virtual machines [1][4]. The service is intended for workloads used by Gemini, internal software teams, and enterprise customers [1]. It arrives as Google expands its infrastructure for agents that can take actions rather than just generate responses [1].

What is Agent Compute Engine

Agent Compute Engine gives each AI agent its own hardware-isolated microVM while keeping credentials outside the sandbox where the agent runs [1]. Google says this combines the separation of VMs with the speed and density usually associated with containers [1]. Traditional VMs can take seconds or minutes to start, suspend or resume, while containers are faster but may lack separation [1]. Agents also spend much of their time waiting between model steps or user instructions, making permanently active VMs expensive [1].

The service is used across four categories inside Google: consumer and enterprise agent tools, software engineering workspaces, large-scale model evaluation and reinforcement learning, and sandbox environments for internal services needing dynamic logic [1].

Security model

A central design assumption is that code inside the sandbox may be untrusted. MicroVMs hold no persistent secrets, OAuth refresh tokens, or cloud service account keys, and they do not access a cloud instance metadata server [1]. Outbound requests route through local and external proxies, with identities and credentials managed off the node [1]. Each agent session receives its own identity and short-lived credentials, injected outside the microVM, and each instance has a cryptographically verified ID [4].

The platform uses Google's Titan and Titanium infrastructure components. Private keys for node and workload attestation are kept outside the guest environment, and storage attachments are mediated by the infrastructure processing unit [1]. Google also separates the orchestration layer — the agent's brain — from the sandbox where code executes, creating an audit trail linking tool calls to activity [1].

Cost, scale and idle optimisation

Google found that most agent sessions are idle most of the time. Agent Compute Engine uses application-aware idle detection so a sandbox can signal when it has entered a waiting state [1]. The host can then checkpoint the sandbox, preserve its state, and reduce active resource use, allowing a single physical node to host far more agent sessions than conventional VMs [1].

The system uses pre-warmed snapshot images and deduplicated memory. Where many agents run the same base image, runtime or browser, the host keeps a single copy of shared static pages in physical memory and allocates additional memory only for data each agent changes [1]. Storage is layered: disposable overlays for temporary OS data, shared read-only volumes for common files, and durable storage only for user working state and saved outputs [1]. For longer tasks, each microVM can use its own Hyperdisk storage to retain workspace state and context between sessions [4].

Complementary open-source runtime: Agent Substrate

Alongside the managed service, Google announced that Agent Substrate, gVisor, and OpenRL have been accepted into the CNCF Sandbox, and Agent Substrate has reached 1.0 and is generally available on GKE [2]. Agent Compute Engine is described as a dedicated microVM compute service with fast hibernation and parallel environment-forking for evaluation and RL pipelines, complementing Agent Substrate [2].

Agent Substrate separates stateful agents from compute by snapshotting idle agents to storage and restoring them in under 500 milliseconds, achieving up to 10x container density according to Google [2]. It runs on any Kubernetes cluster and supports standard OCI containers, working with Agent Development Kit agents, stateful coding environments, and productivity agents [2].

Governance for enterprise agents

Google Cloud also expanded agent governance tools to cover agents across its services and other environments [4]. The framework includes Agent Gateway for policy enforcement and traffic control, Agent Registry for approved agents and tools, Agent Identity for secure identities, and observability based on OpenTelemetry [4]. New governance functions due in preview include token and cost controls, model routing policies, extension bindings, and delegated user authorisation through identity partners including Ping Identity, SailPoint and Saviynt [4].

What it means for developers

For developers building autonomous coding agents, Agent Compute Engine targets the need for persistent Linux environments with root privileges, custom compilers, test runners, and persistent git workspaces [1]. Google DeepMind's Clement Farabet said the service provides "instant snapshot-and-fork primitives and rapid resume speeds" for long-horizon software engineering [1]. The design lets agents branch into parallel experiments quickly, which is relevant for multi-step development work rather than single code suggestions [1].

Agent Compute Engine may reduce the cost of idle agent sessions by checkpointing instead of keeping VMs running. Its security model moves secrets and policy decisions outside the sandbox, which matters for teams running untrusted model-generated code. The open-source Agent Substrate offers a Kubernetes-based alternative for platform teams wanting customisation or multi-cloud portability [4]. No pricing has been disclosed in the sources.

Sources

  1. Google Cloud unveils Agent Compute Engine for AI agents
  2. Open infrastructure for autonomous agents comes to CNCF | Google Cloud Blog
  3. Google Cloud launches AI agent engine & governance
  • google cloud
  • ai agents
  • microvm
  • agent substrate
  • developer tools

Researched and drafted with AI from the sources above.

Learn to build with AI

Hands-on tutorials on Cursor, Claude Code, AI agents and more.