Google on November 19, 2026 added computer use to the Gemini app for Mac, so the assistant can carry out multi-step tasks across local files, desktop applications and websites. Michael Friedman, a Group Product Manager for the Gemini app, announced the feature, which Google describes as handling the clicking, typing, menu navigation and form filling that normally takes dozens of manual steps.
A user describes the job in plain language and Gemini works through it inside an isolated background window, with a live preview shown in the chat. Because the agent has its own window, the user's main screen and cursor stay free, and they can keep typing, designing or taking calls while the task runs. Google lists several examples: batch color-grading photos in design software, extracting weights, dimensions and addresses from local PDFs to complete shipping quotes, pulling metrics and screenshots from dashboards across regions into a Google Slides deck, and analyzing clothing photos from a folder to research prices and stage resale listings.
Google frames control as the central design point. People choose which apps Gemini may access, watch each step live, pause at any time and approve key actions before they happen. TechCrunch notes that, as with other computer-use agents, users can pause, stop and approve actions, and describes the approach as familiar from competing products.
Availability is narrow at launch. The feature is rolling out in English in the United States, requires the Gemini app for Mac on an Apple Silicon computer running macOS Sequoia 15.0 or later, and is limited to personal Google accounts for users aged 18 and over who subscribe to Google AI Plus, Pro or Ultra. Google did not say when it would reach other countries or platforms, according to TechCrunch.
Independent coverage adds context. TechCrunch's Sarah Perez wrote that Google is somewhat late, since OpenAI, Anthropic and Meta have offered agentic features in their desktop apps for some time, and noted that Google only this week widened the rollout of Gemini Tasks. The Verge's Jay Peters pointed out that the launch comes ahead of new Apple restrictions on apps requesting full disk access on Mac, motivated by the risks of AI agents using computer control. A preview of those settings appeared in this week's macOS 27.3 beta.
For users, the practical test will be reliability on real workflows and how the permission model interacts with Apple's coming controls. The announcement is a product description, and Google has published no accuracy, speed or failure-rate data for the feature.