Gemini Takes the Wheel on Mac: Google’s AI Agent Now Clicks, Types and Fills Forms for You

Google has quietly crossed a line that many users have been waiting for — and some have been dreading. On November 19, 2026, the company announced that its “computer use” capability is now live inside the Gemini app for Mac, allowing the assistant to operate the machine on your behalf: clicking, typing, navigating menus and filling out forms to complete multi-step tasks described in plain language. The move turns Gemini from a conversational chatbot into something closer to a digital intern — one with access to your files, your desktop apps and the web.

  • Agentic automation arrives on macOS: Gemini can now execute multi-step workflows across local files, native applications and websites, all from a single prompt.
  • Isolated execution with live preview: The agent runs in a separate background window, leaving your screen and cursor free while you watch every step in a live preview inside the chat.
  • Limited rollout, strict controls: The feature is available only in English and the United States, requires an Apple Silicon Mac running macOS Sequoia 15.0 or later, and demands a Google AI Plus, Pro or Ultra subscription.

What Gemini Can Actually Do on Your Mac

According to Google, the feature is designed for the kind of tedious, repetitive work that eats up entire afternoons. The company’s examples are ambitious: correcting the color of a batch of photos in a design program, assembling complex 3D scenes, extracting weights, dimensions and addresses from multiple local PDFs to fill out shipping quotes with dozens of fields, pulling metrics and screenshots from live dashboards across different regions to build a Google Slides presentation, or analyzing photos of clothing in a folder, researching prices and preparing resale listings.

To trigger the agent, users type @Computer use in their request. The assistant then works inside an isolated window, with a live preview visible in the chat. Your main screen and cursor remain untouched, so you can keep working while Gemini handles the clicks and keystrokes in the background.

The launch also brings new desktop connectors to the Gemini app, including Blender and Autodesk Fusion, which join existing integrations such as Git and Obsidian. Gemini’s Mac tasks could already access local files alongside connected apps like Gmail and Google Calendar; the new capability extends that reach to the graphical interface of virtually any installed program.

Safety Rails and the Limits of Delegation

Google is keen to stress that the user stays in control. You choose which applications Gemini can access, you can watch every step in real time, pause at any moment, and you must approve key actions before they happen. According to reports, the assistant stops at logins and payments and asks for confirmation on sensitive operations.

“You choose which apps Gemini can access, you can see every step live, pause whenever you want, and you have to approve key actions before they happen.”

Even so, Google’s own help documentation warns that Gemini can make mistakes. The assistant may permanently delete files or share them with third parties, although it is designed to ask permission first. Google recommends not connecting files you would not want it to touch, or that are too sensitive to risk losing. Gemini only sees the folders you connect, and when it modifies files it creates temporary backups that are deleted when a new task starts or after 24 hours. The help center also cautions against adding public cloud storage to connected folders, since the assistant could inadvertently share data there.

The timing is notable. Apple is preparing macOS 27.3, an update that will introduce warnings about applications accessing user data, allow finer-grained limits on app and folder access, and include an indicator showing when one application is controlling another. Users who grant these permissions now will not have to re-grant them when that update arrives, according to reports.

Availability, Pricing and the Road Ahead

For now, the feature is geographically constrained. It is rolling out in English and only in the United States, for personal Google accounts belonging to users over 18 who subscribe to Google AI Plus, Pro or Ultra. Plans start at $4.99 per month, according to MacRumors. The app is available for download at gemini.google/mac and requires a Mac with an Apple Silicon chip and macOS Sequoia 15.0 or later. Google has not said when the feature will expand to other markets or platforms, and it has not named the specific model powering the capability.

The launch places Google behind competitors such as OpenAI, Anthropic and Meta, which have already been offering agentic features in their desktop applications for some time. Google only recently announced the broader rollout of Gemini’s “Tasks” feature, which lets the AI sign into websites, make purchases or book appointments, and it has begun admitting early testers to its revamped “CC” AI agent announced in September.

For users in Europe and the UK, the wait continues — the feature is not available there, and no timeline has been given. When it does arrive, the sensible approach will be to start with a small, low-stakes task and watch every step before delegating more. The productivity gains are real, but so is the habit of clicking “approve” a little too quickly. An agent with permission to click and type on your computer is also an agent that can err — or be misled — with tangible consequences for your files and accounts. The tool is impressive; the discipline it demands is what will ultimately determine whether it becomes a trusted assistant or a cautionary tale.

Sources & References