THE SIGNAL · November 20, 2026
Google says its Gemini app for macOS is rolling out a new “computer use” capability that can carry out multistep work across a user’s local files, desktop applications, and the web. The release moves Gemini beyond a chat interface: the company says users can authorize access to selected apps and watch the agent’s activity through a live preview in the chat.
What happened
According to The Verge, Gemini for macOS can now help complete work involving files on a Mac, desktop apps, and web services. Google describes potential uses including filling out forms, preparing presentations, speeding up workflows, and creating listings for items for sale.
The feature is rolling out in English to users in the United States. Google says users will be able to decide which apps Gemini can access. It also says Gemini will operate with a separate background window that provides a live preview within the conversation, allowing users to follow what it is doing.
The announcement is a notable expansion of what an AI assistant can do on a personal computer. Rather than simply producing text, an assistant with computer-use permissions can potentially navigate among the tools and information involved in a task. That makes the boundaries of app access, the user’s ability to monitor activity, and the clarity of permissions central parts of the product experience.
Why it matters
Computer-use features are becoming an important front in the competition among AI companies. The Verge notes that other AI products can already take control of computers through Mac applications, and that Google is introducing Gemini’s capability as Apple is reportedly working on an agentic assistant of its own.
For people who use a Mac for administrative or repetitive work, the practical appeal is straightforward: tasks often require moving information between documents, browser tabs, local folders, and applications. An assistant that can work across those places could reduce some of that manual switching. But the same breadth is what makes the feature different from a conventional chatbot. A system that can interact with an app may encounter personal files, account information, or material that is not meant for every application on a device.
Google’s stated app-by-app access choice is therefore more than a setup detail. It is the primary control described in the available information. The live preview may also give users a way to supervise an agent as it works, rather than receiving only a final answer after the fact. The packet does not specify how Gemini handles mistakes, what it can do once access is granted, or what additional limits apply to sensitive data and actions.
That uncertainty does not mean the feature is unsafe; it means users should understand the permissions they are granting and begin with tasks whose results they can readily review. In practice, the value of computer use will depend not just on whether an agent can complete a sequence of clicks, but on whether people can tell what it is doing and intervene when necessary.
What to watch
- Rollout: Google says the capability is currently rolling out in English for U.S. users. Availability beyond that group has not been detailed in the packet.
- Permission controls: How clearly Gemini presents app-access choices, and how easily users can change them, will shape how much practical control users retain.
- Supervision: Google’s live-preview approach could become an important test of whether agent activity is understandable while it is happening.
- Use cases: The early examples—forms, presentations, workflow tasks, and sale listings—suggest Google is aiming at routine desktop work rather than a single specialized job.


Leave a Reply