Google’s Gemini Can Now Operate Mac Apps to Finish Tasks for Users
Google is rolling out computer use in the Gemini app for Mac, so the AI assistant can click, type and fill out forms in a background window for paying US users.
Google has started letting its Gemini assistant take control of Mac software to carry out multi-step chores for users. A new “computer use” feature in the Gemini app for macOS can click, type, navigate menus and fill out forms across local files, desktop programs and websites. It works in a separate background window, so the person at the keyboard can keep working. The feature began rolling out on Thursday to paying subscribers in the United States, Google said in a blog post.
What Google announced
The announcement came in a post by Michael Friedman, a group product manager for the Gemini app. He described a tool meant for tedious jobs that are spread across several programs. Users describe the task in plain language, and Gemini “handles the clicking, typing, menus, and forms,” the company wrote.
According to Google and to reports by the tech sites 9to5Google, Engadget and The Next Web, the agent works by taking screenshots to see what is on screen and then acting on what it sees. It does not take over the user’s main display or cursor. Gemini runs inside an isolated window and shows a live preview of its progress in the chat. 9to5Google reported that users start the feature by adding “@Computer use” to a prompt. It also reported that Google has added desktop connectors for programs such as Blender, the open-source 3D modeling software, and Autodesk Fusion, a design and engineering tool.
Google gave several examples of what the agent can do:
- Batch color-grading photos in design software or setting up complex 3D scenes.
- Pulling weights, dimensions and addresses from local PDF files to fill out multi-field shipping quote forms.
- Collecting metrics and screenshots from live dashboards and building a presentation in Google Slides.
- Analyzing photos of clothing stored in a folder, researching prices and staging listings on resale sites. Engadget named eBay as one such platform.
Who can use it, and the safeguards
The feature is rolling out in English in the U.S. to subscribers of Google AI Plus, Pro and Ultra, the company’s paid AI plans. It is limited to personal Google accounts belonging to users 18 and older. It requires a Mac with Apple Silicon running macOS Sequoia 15.0 or later. The Next Web reported that, like Google’s earlier Gemini Spark agent, the feature is not available in Europe or the U.K.
Google says users choose which apps Gemini may access, can watch each step as it happens, can pause the agent at any time and must approve key actions before they are carried out. 9to5Google reported that Gemini will stop for logins and payments and ask for confirmation before sensitive actions.
The company is also open about the risks. The Next Web reported that Google warns Gemini can make mistakes, including permanently deleting files or sharing them with others. The agent is designed to ask permission first and can only see folders the user has connected. Backups of changed files are deleted after 24 hours or when a new task begins, according to the outlet.
Background: the race to put AI agents on the desktop
The launch builds on agent tools Google has been adding to its Mac app since it arrived in April. 9to5Google said the app’s existing Task feature, formerly called Gemini Spark, could work with local files only together with connected Google services such as Gmail and Google Calendar. The new mode extends Gemini’s reach to third-party desktop programs and the open web.
Reports differ slightly on the feature’s history. Engadget wrote that Gemini had already been able to handle repetitive tasks for PC users for some time and that Mac users were now catching up. Google’s announcement does not mention Windows. The Next Web noted that Google first offered computer use to developers through its Gemini 3.5 Flash model in June.
The move puts Google deeper into a crowded contest among AI companies to build “agents,” which are assistants that act on a user’s behalf instead of only answering questions. Rivals including OpenAI and Anthropic have released their own tools that can operate computers. Apple is also hiring engineers to build its own agentic assistant, The Verge reported this week, citing a job listing first spotted by Bloomberg. The Next Web said Google is also building Gemini into the operating system of its Googlebook laptops.
What comes next
Google has not said when computer use will reach more countries, more languages or its free tier. The rollout also comes amid growing concern about how much access AI agents should have to personal data and devices. This week the messaging app Signal said it would limit screenshots by default to keep operating-system AI agents from reading users’ chats. For now, Google is relying on user permissions, live previews and approval prompts to keep the agent in check, while acknowledging that it can still get things wrong.

