News from the IT community on October 23: Google announced yesterday (October 22) the launch of an Agent Computing Engine (referred to as ACE), which is a computing service specifically designed for AI agents. The goal is to provide each agent with a secure, durable, and scalable "exclusive computer."
According to the official announcement, the ACE service will provide each AI intelligent entity with an independent hardware-isolated microVM (micro virtual machine), allowing them to securely execute code, call tools, control browsers, and complete multi-step tasks.
IT Home援引博文介绍，不同于传统虚拟机或容器，ACE 针对智能体的生命周期特征专门优化，可快速启动，具备独立客户内核和根权限，同时通过硬件隔离的Cloud Hypervisor边界保障安全。谷歌表示，沙箱内不会保存凭据，以降低智能体被滥用或数据泄露的风险。
The highlight of the ACE service lies in its "persistent and pauseable" feature. When the agent is idle, its computing environment can be automatically suspended to free up resources; when work needs to continue, the file and memory states can be quickly restored.
Google claims that in tests with 200,000 agents, the time taken to pause and resume is less than 500 milliseconds, and by sharing memory and using snapshots, the number of agents that a single server can support can be significantly increased.
This service was initially used to support the model evaluation and reinforcement learning processes of Gemini, Gemini, Antigravity platforms, as well as within Google. In the future, developers will be able to use ACE through Google Cloud to control the images, identities, network access, and lifecycle of intelligent agent environments on their own, while Google will be responsible for the underlying resource scheduling and scaling.
