The agent works in an isolated window with a live preview, and pauses for logins, payments and other sensitive actions by default.
· 4 min read
Google has added computer use to its Gemini app for Mac, letting the assistant click, type, open menus and fill out forms across local files, desktop apps and websites. The feature began rolling out on November 19 in English in the United States, and it is limited to subscribers of Google's paid AI plans, according to Google's announcement.
How it works
Gemini runs the task inside an isolated background window, with a live preview shown in the chat. That design leaves the user's main screen and cursor free, so you can keep working while the agent proceeds. Google says users choose which apps Gemini can access, watch every step live, pause at any time and approve key actions before they happen. By default the agent pauses for logins, payments and other sensitive actions, according to coverage of the launch.
To start a task, you include @Computer use in the prompt and describe what you want done, 9to5Google reported, as relayed by Android Authority. Google's blog suggests uses such as batch colour-grading photos in design software, extracting weights, dimensions and addresses from local PDFs to fill in shipping quote forms, building a presentation in Google Slides from live dashboard data, and analysing clothing photos from a local folder to stage resale listings.
Who can use it
Computer use requires a Mac with an Apple Silicon chip running macOS Sequoia 15.0 or later, a personal Google account held by someone aged 18 or over, and a Google AI Plus, Pro or Ultra subscription. The Gemini app for Mac itself is a free download at gemini.google/mac. The launch is limited to English and to the United States, and Google has given no timeline for other countries or for Windows or Chromebooks.
The entry price is modest compared with other agent products. PPC Land, citing a report from October, listed the plans as $4.99 a month for AI Plus, $19.99 for AI Pro and $99.99 or $199.99 for AI Ultra. Google has not changed those figures with this launch, and the plan details may differ by market.
What it means for people in India
Computer use is not available in India at launch, because it is limited to the US. That contrasts with the Tasks feature in the Gemini app, which went free in India and five other countries earlier this week, though calling businesses and making payments stayed US-only. Our report on Gemini Tasks going free in India explains that rollout. Indian Mac owners who want to try computer use would need to wait for a wider release.
Where it fits in the agent race
Google arrives at computer use after OpenAI, Anthropic and Meta, AI Weekly noted, and the Mac launch comes only two days after Google rebranded its Gemini Spark agent as Tasks and removed its paywall in six countries. The company also released the Gemini 4 Neon model last week, which we covered in our story on the model's launch and pricing, so the agent feature sits on top of a quickly changing product line.
The timing also overlaps with Apple's own plans. Bloomberg reported the same day that Apple is building an autonomous agent of its own, a story we cover in our report on Apple's AI agent job listings. Bloomberg's Mark Gurman said on X that Apple needs to speed up, because Google now offers computer use on the Mac alongside ChatGPT and Meta's Muse.
Risks and caveats
Handing an AI the keyboard and mouse is a different level of trust from asking it a question. Google acknowledges that the feature may make mistakes, which is why it asks for approval before key actions and pauses for sensitive steps. AppleInsider's headline on the launch put the caution bluntly: Gemini can now control Mac apps, but you still should not let it. That is an editorial view, not a finding, though it captures the practical advice to supervise early tasks.
There are also questions about how apps will respond. Apple has been adding permission screens and tougher controls in macOS 27.3 beta, as noted in our coverage of full disk access and agents, so third-party agents may meet more friction on the Mac in future releases. Signal has gone further and begun blocking screenshots by default to keep assistants from reading chats, a move we explain in our report on Signal's new protection.
What to try first
For early users, the safest starting point is repetitive, low-stakes work: renaming or sorting files, extracting figures from documents into a form, or compiling a slide deck from public data. Tasks involving payments, passwords or sensitive personal data are better handled manually until the behaviour is well understood. Google says you can stop a run at any moment, and the live preview makes it easy to see when the agent has misread a screen.
Google has not said how it will expand the feature. The most likely next steps are more languages and regions, and eventually a version for Windows, though the company has indicated no plans for either so far.
Published November 21, 2026. Gadgets365 will update this article as more information becomes available.
Editor
Pravin KankanePravin Kankane is the editor of Gadgets365, an independent technology publication based in Pune, India. He covers smartphones, AI, Apple, Android and consumer tech launches in India.


