Announced at AGNTCon + MCPCon: the new Agent Compute Engine plus expanded Gemini for business agent governance deliver the control you need to run, govern, and scale your agents.
Learn more about Agent Compute Engine and expanded agent governance → goo.gle/4yFN3yT
Learn more about Agent Compute Engine and expanded agent governance → goo.gle/4yFN3yT
Oct 22, 2026 · 3:06 PM UTC
37
44
305
36,060
@googlecloud The handoff between runtime and governance is the interesting part here. Can an agent’s permissions be revoked during a running task, with the blocked tool call and approval history visible in the same audit trail?
2
0
1
1,054
@RyanJ_KOL Hi there! We appreciate you bringing this to our attention. To protect your personal privacy and account security, please send us a DM so we can look into this for you. We are here to support you and look forward to helping. -JB x.com/messages/compo…
0
0
1
703
@googlecloud Agent Engine was already complex and confusing, thank you for making it more complex and confusing :D
0
0
1
115
agents with their own computers will be a big deal, but a computer per agent gets expensive. google's answer: pause the computer while the agent is thinking.
its new agent compute engine gives each gemini agent its own microvm. when the agent finishes a tool call and starts human, the vm goes idle and resumes from a snapshot later. each one also keeps its own disk, so the workspace survives between sessions.
it already runs gemini's agent, and customers get a preview in the coming months. the open-source kubernetes version, agent substrate, works with claude code, openclaw and hermes, and google says it packs up to 10x more onto the same compute.
0
0
0
248
▲ Google Cloud just announced Agent Compute Engine, a compute service that gives every AI agent its own hardware-isolated microVM with zero credentials inside.
The key idea is splitting each agent into a harness (the brain) and a sandbox (the hands). The sandbox has no network ffic leaves over vsock, and Agent Gateway swaps in the real credential off-node, so a prompt-injected agent finds nothing to steal.
Idle agents are checkpointed to zero active resources and resumed from snapshots, with copy-on-write memory shared across microVMs. Its open-source sibling, Agent Substrate on GKE, reports 500 activations/s at under 500 ms suspend/resume across 200,000 agents. It is a trade-off though: Google publishes no boot time, density or price for Agent Compute Engine itself, and it is not available to customers yet.
It already powers Gemini agents at Google and enters preview in the coming months. Agent Substrate is open source and available on GKE for non-production use today......
Full analysis: marktechpost.com/2026/10/22/goo…
Docs: cloud.google.com/blog/topics/ai…
Repo: github.com/agent-substrat…
@googlecloud @GoogleAI @googleaidevs
0
0
1
107
@googlecloud 谷歌这个判断我认：容器配不上agent。每个agent独占microVM、沙箱里零凭证、密钥在网关外注入，闲下来快照冻结不烧钱——这才是生产级agent该有的房子。注意20万agent每秒激活500次是开源版Agent Substrate的数，别张冠李戴。你给agent开权限时，是先给钥匙还是先装防盗门？
0
0
0
21
@googlecloud Once an agent wants to serve or fine-tune its own model, the microVM isn't enough. Who's giving it GPUs by the second?
0
0
0
26
@googlecloud agent compute engine sounds like it’ll just add another bill line i’ll avoid reading.
0
0
0
70
@googlecloud agents finally getting a proper home to run in, about time
0
0
1
119
@googlecloud Whatever cloud you're on, each agent should have its own login, and its logs should live somewhere it can't touch.
0
0
1
94
