Google Antigravity adds cloud remote agents that work with PC off

Google has added "remote agents" to Antigravity, its development environment built around AI agents, so that the agents keep working in the cloud. The Japanese outlet Publickey covered the feature on 12 November 2026, and Google said through its official account on 6 November 2026 that it had been rolling the feature out that week.

A few terms first. An AI agent is a program that receives an instruction and carries out several steps on its own judgment. A virtual machine (VM) is an independent computer hosted in the cloud, separated from other working environments. A pull request is the mechanism for asking a team to review code changes before they are merged.

What does a remote agent actually do?

According to Publickey, in a typical development setup the code sits on the local machine's file system, and a programmer or an AI agent writes it, builds it and tests it there. The new feature runs that work in an isolated remote cloud environment managed by Google. The agent therefore keeps working whatever the state of the PC in front of the user.

In an article dated 7 November, the US outlet Thurrott.com explains that remote agents can handle long-running tasks even when Antigravity or the browser is not open, or when the computer is offline or asleep. Publickey likewise reports that after a lunch break, a business trip or the end of the working day, the agent carries on in the cloud even if the PC is switched off.

Your tasks run in independent cloud virtual machines. Fire off a task, step away for coffee, unblock permissions from your phone, and come back to passed tests and a clean pull request.

When a confirmation or an instruction is needed, Publickey says the user can reach the agent working in the remote environment at any time, from a smartphone for example. Google's post also says permissions can be unblocked from a phone while the user is away. The design appears to assume that people no longer sit next to the agent and step in only when a decision is needed.

Which code and files can be handed to the agent?

According to Publickey, the code and related files a remote agent works on can be set up by pointing to a GitHub repository, by choosing files or folders from Google Drive, or by uploading files from the user's device. Unlike the older way of working, where the agent reads a local folder directly, the work material has to be handed over to the cloud side first.

Who can use it, and under what conditions?

According to Publickey, the feature is currently available to subscribers of Google AI Pro and Google AI Ultra who sign in with a personal Google account. Google's post says it is available on all paid consumer plans. Publickey's article does not say how business accounts or contracts are handled.

Google says the rollout began that week, so it is worth remembering that availability may still be spreading. Usage limits, running time and any extra fees are not mentioned in the articles read for this piece. AI coding tools change quickly and the eligible plans may change too, so the current conditions should be checked in Google's official information before adoption.

What else was updated at the same time?

According to Thurrott.com, remote agents are part of Antigravity version 2.29, available for Windows, Mac and Linux. The same update brings a voice mode. Built on the Gemini Live model, it lets users plan, steer code changes and approve permissions by voice. It supports conversations in several languages, interruptions mid-answer, and a choice of voice and playback speed.

The interface changed as well. The outlet reports that on the Conversation page the prompt box moved to the bottom of the screen, fonts were updated throughout, and the agent picker moved into the model menu. There are more changes, and the full list is on the Antigravity site. The app can be downloaded for Windows 11, Windows 11 on Arm, Linux, Mac and Googlebook.

What this changes for a company in Japan

The main change is that developers no longer need to watch the AI work in front of a screen. Handing over a task in the evening and reviewing tested changes the next morning becomes realistic at an individual level. For small development teams short of people, or companies maintaining internal tools with a handful of staff, it may be a reason to rethink how working time is used.

There are points to weigh as well. Because the code and files being worked on are entrusted to the cloud, a company has to decide internally which information may leave its premises. The current offer centres on consumer plans, and what business-grade controls will look like cannot be told from these articles. A realistic start is to try it on low-sensitivity projects and to make human review of every agent-made pull request a fixed rule.

For guidance on rolling out agents and keeping them under control, see also agent governance.