Contents

Google Antigravity now lets AI agents run development tasks on independent cloud virtual machines, allowing them to keep working after you shut down your computer. As of November 13, 2026, remote agents are available to Google AI Pro and Google AI Ultra subscribers signed in with personal Google accounts.

Key Points#

  • Remote agents run on cloud virtual machines and can continue tasks independently of your PC.
  • Project materials can be provided through GitHub, Google Drive, or file uploads from your device.
  • You can access agents from your smartphone when they need permission approvals or further instructions.
  • Google targets Google AI Pro and Google AI Ultra subscribers using personal accounts.

What Runs in the Cloud#

Remote agents move task execution to an independent cloud environment managed by Google. Your computer does not need to stay powered on or connected; the agent works on a dedicated virtual machine.

Google’s November 6 announcement describes a workflow in which you submit a task, step away, approve permissions from your smartphone, and return to review test results and a pull request. This is an intended use case, not a guarantee that every task will succeed or produce changes ready to merge.

The practical change is freedom from dependence on your device. Closing your laptop no longer means ending the agent’s work, though tasks may pause when input is needed.

How to Prepare a Task#

There are several ways to provide working materials: specify a GitHub repository, select files or folders from Google Drive, or upload files from your device. Choose sources that contain the code and supporting information needed for your request.

Before starting a remote task, prepare these four things.

1. Identify the repository and relevant files or folders. 2. Describe one narrowly scoped change, such as fixing a reproducible bug. 3. Specify the expected behavior and the checks that must pass. 4. Explain which decisions require your approval.

A small issue with clear reproduction steps and an existing test command makes a good first request. Broad requests such as “improve the application” leave too many unresolved questions in work you intend to review later.

What Changes Outside Working Hours#

Remote execution lets you start a task before lunch, a commute, or the end of the workday and keep it running while your PC is off. This assumes the project can run in the cloud environment, but it is useful for work that involves waiting for builds or tests.

Your smartphone becomes a way to unblock work. According to Google, users can approve permissions remotely, and the provided report says they can access agents when clarification or further instructions are needed.

However, not every task can run entirely unattended. When an agent encounters a decision that needs your answer, progress may depend on when you respond. Write tasks so routine choices are clear before you step away, and decide whether you will respond outside working hours.

What to Check Before Relying on It#

Eligibility and operational limits need to be checked separately. The provided report identifies personal Google accounts with Google AI Pro or Google AI Ultra subscriptions as eligible; availability for managed work accounts has not been confirmed. It also does not specify task limits, maximum runtime, supported regions, or additional fees. Check the latest terms before planning substantial work.

An independent virtual machine isolates the execution environment, but that does not mean every repository is suitable for upload. For work projects, confirm that your organization permits this service to process the code and related files. Before assigning tasks that require credentials or private dependencies, also check how those will be made available.

Review the results as you would any proposed code change. Passing tests provide useful evidence, but they do not prove that the agent interpreted the requirements correctly. Check the changes and test output before merging a pull request.

Next Steps#

Give a remote agent one small, reproducible bug to fix, along with an explicit test command, and review the changes and test output before assigning larger tasks.

Source: https://www.publickey1.jp/blog/26/google_antigravityaiai.html

Contents