Google today began rolling out computer use in the Gemini app for Mac, a feature that lets its assistant work through desktop applications, local files and websites on a user's behalf from inside an isolated background window. The rollout is limited at launch to English in the United States and to Google AI Plus, Pro and Ultra subscribers, according to the announcement by Michael Friedman, Group Product Manager for the Gemini app, published on Google's blog at 17:00 UTC.

In Short

Google's Gemini app for Mac can now click, type and fill in forms across desktop apps, local files and websites while you keep using your own screen and cursor. It is reaching paying Google AI subscribers in the US first, on Apple Silicon Macs, two days after Google made its background task agent free in six countries. For marketers and site owners, it adds another AI agent that works through screens rather than APIs, and Google has not said how its web activity will identify itself to the sites it visits.

What Google announced

The post is short, at about 230 words, and leads with time saved. "Starting today, computer use in the Gemini app for Mac helps you get multi-step tasks done across your local files, desktop apps, and the web," according to Google. The work it targets is "the repetitive work that normally takes dozens of clicks." Instructions are given in plain language: "Just describe what you need, and Gemini handles the clicking, typing, menus, and forms for you."

That description places the feature squarely in the category of a computer use agent: software that operates a computer through its graphical interface, reading what is on the screen and acting on it, rather than calling an application's API. Google has been building in this direction for some time. Its Gemini 2.5 Computer Use model was released to developers on October 7, 2025, and Project Mariner, its experimental browser agent, was described by PPC Land in March as Google's agent for completing tasks inside a web browser. What changes today is the surface. The capability now sits inside a consumer app on a personal computer and reaches beyond the browser into any desktop program a user allows.

The design choice Google highlights is where the work happens. "Gemini works inside an isolated background window with a live preview right in your chat," the post states. "Your main screen and cursor remain completely free, so you can keep typing, designing, or taking calls without interruption." The arrangement means the agent and the person are not competing for one mouse pointer: Google's version runs alongside the user rather than in place of the user.

Controls are described in a single sentence. According to Google, "you choose which apps Gemini can access, watch every step live, pause at any time, and approve key actions before they happen." The app-selection control works as a form of least-privilege access: the agent reaches the applications a user names rather than the whole machine. The approval step is a human-in-the-loop checkpoint, although the post does not define which actions count as "key."

Four tasks Google chose to show

Google listed four examples under the line "Here are a few ways you can start trying it today." The wording below is Google's.

Use case (Google's label) Google's description Material involved
Speed up creative workflows "Batch color-grade photos in your favorite design software or set up complex 3D scenes." Design and 3D software
Skip manual form entry "Extract weights, dimensions, and addresses from local PDFs to fill out multi-box shipping quotes." Local PDFs, shipping quote forms
Turn live data into ready-to-present decks "Pull metrics and screenshots from live dashboards across multiple regions and build a presentation in Google Slides." Live dashboards, Google Slides
Batch-list items for sale "Analyze clothing photos from a local folder, research pricing, log item details, and stage listings on resale platforms." Local photo folder, web research, resale platforms

The hero image in the post makes the first case concrete. It shows a prompt that reads "Could you take my kid's butterfly sketch, model it and rig it for me in Blender? @Computer use", followed by a "Thinking it through" panel listing two steps, "Check workspace directory and Blender installation" and "Creating and texturing butterfly model", and then a live view of Blender with a three-dimensional butterfly on the canvas.

At least two of the four examples involve websites, and three describe the kind of work marketing teams and online sellers do every week. Shipping quotes and resale listings are e-commerce chores. A deck built from "live dashboards across multiple regions" describes the monthly reporting cycle that agency account teams know well.

How it works, according to Google

Google's post and its footnote, together with reporting by 9to5Google, describe the feature in seven steps.

  1. Hardware and system. The footnote reads: "Requires an Apple Silicon Mac running macOS Sequoia 15.0+." Intel-based Macs are excluded.
  2. Installation. The post points to gemini.google/mac to download the Gemini app for Mac. Google's help page for the app lists 8 GB of RAM and 200 MB of free disk space for installation.
  3. Account and plan. The footnote continues: "Available for personal Google accounts 18+, available to Google AI Plus, Pro and Ultra users." Work and school accounts are therefore outside the rollout, even though the Mac app itself accepts them where an administrator has enabled Gemini, according to Google's help page.
  4. Invocation. A request is typed into the chat with "@Computer use" added, as the prompt in Google's hero image shows. According to 9to5Google, adding that tag is how the feature is called.
  5. Choosing apps. The user selects which applications Gemini may access. According to 9to5Google, new Desktop connectors, among them Blender and Autodesk Fusion, appear in the app's Customize > Connectors menu.
  6. Execution. Gemini works in the isolated background window, with a live preview in the chat, while the main screen and cursor stay with the user.
  7. Oversight. The user can watch each step, pause at any time and approve key actions before they run. According to 9to5Google, Gemini also pauses for logins and payments and asks for confirmation before sensitive actions.

Who can use it

The eligibility rules sit across three Google documents, and they do not match one another. Computer use is narrower than both the Mac app it lives in and the task agent that runs in the same app.

Requirement Computer use (Google post, today) Gemini app for Mac (Google Help and product page) Tasks in Gemini (Google Help)
Market and language English, United States Free in all languages and countries where the Gemini app is supported Free in Australia, Canada, India, Japan, South Korea and the US; Pro or Ultra elsewhere, except the EEA, Nigeria, Switzerland and the UK
Hardware and OS Apple Silicon, macOS Sequoia 15.0 or later Apple Silicon, macOS Sequoia 15.0 or later, 8 GB RAM, 200 MB disk Mobile app, Mac app or web app
Account Personal Google accounts Personal, or work and school where enabled by an administrator Personal Google Account only
Age 18+ Not stated on the help page 18 or over
Plan Google AI Plus, Pro or Ultra No cost No subscription in the six countries listed
Other Not stated Not stated History must be on

Plan prices reported by PPC Land on October 4 put AI Plus at $4.99 a month, AI Pro at $19.99 and AI Ultra at $99.99 or $199.99. On that basis, the lowest monthly price at which a US user can reach computer use is $4.99. No source checked for this article states that free accounts will receive the feature. TechCrunch reported that Google did not say when it would expand to other markets or platforms.

The tiering is the most telling detail. Since November 17, Google's task agent on the Mac has been free in the US. Computer use, which lets the same app operate other programs, starts at a paid tier. Google has split its desktop agent into a free layer that handles files and connected services and a paid layer that drives the screen.

From Spark to computer use in 142 days

The Mac has been Google's testing ground for desktop agents for most of the year. On June 30, Google brought its Gemini Spark agent to macOS for US Google AI Ultra subscribers aged 18 and over, with access limited to the files a user gave it permission to use. A month later, on July 29, the company gave Gemini for Mac voice control over any open app, with dictation into any application through a long press of the Fn key. On July 30, Spark expanded to AI Pro subscribers in more than 160 additional countries but remained unavailable in the European Economic Area, the UK, Switzerland and Nigeria.

The Windows side followed on September 10, when Gemini became an installable Windows program opened with Alt + Space. Today's announcement covers only the Mac.

The autumn brought price movement. On October 13, Google added a tasks mode that keeps working while users are away, for AI Pro and Ultra subscribers. On November 17, the same task mode went free in six countries on Gemini 4 Hydrogen, and The Verge reported the same day that the product formerly known as Spark was being renamed Tasks, a name that Google's own to-do app has carried since 2018. In its report on computer use today, 9to5Google described the existing Task feature as the former Gemini Spark. Google's Mac product page still described Spark today as rolling out first to AI Ultra users aged 18 and over, a description its help pages no longer match.

From the Ultra-only Spark launch on June 30 to today is 142 days, by PPC Land's count. In that time the entry price for a Google agent on the Mac has moved from the top consumer tier to zero for file-based tasks in the US, and to the bottom paid tier for screen control.

Date Google agent release Access at launch
June 30, 2026 Gemini Spark on Mac (beta) AI Ultra, US, 18+
July 14, 2026 Chrome auto browse AI Pro and Ultra, US
July 30, 2026 Spark expansion AI Pro, 160+ more countries, not EEA, UK, Switzerland, Nigeria
October 13, 2026 Tasks mode in the Gemini app AI Pro and Ultra
November 10, 2026 Android task automation in three more countries No subscription; India, Japan, South Korea, UK, US
November 17, 2026 Tasks mode free Six countries, Gemini 4 Hydrogen
November 19, 2026 Computer use on Mac AI Plus, Pro and Ultra; US English; personal accounts 18+

Three ways to run an agent

Google now runs its assistants' actions in three different places, and the differences matter for anyone trying to understand where an agent's activity comes from.

On Android, Google's task automation for food and ride apps runs inside an on-device virtual phone environment, which PPC Land described on November 12 as a computer use agent running on the handset. In the cloud, Google said in its November 17 task mode post that tasks run on Google Cloud's Agent Compute Engine, in a hardware-isolated virtual machine dedicated to each account. On the Mac, computer use works in an isolated window on the user's own computer, by Google's description.

Browser-based agents have taken a different path. OpenAI's update to its Chrome extension, which PPC Land covered on October 16, let ChatGPT control browser tabs to fill forms and send messages. Google's own Chrome auto browse lets agents book travel and fill in forms for AI Pro and Ultra subscribers in the US. Both are versions of an agentic browser, with the agent confined to web pages. Anthropic's Cowork, which arrived in January as a research preview for Claude Max subscribers on macOS, works through folders a user grants. Today's Gemini feature reaches further than either approach, into desktop software such as Blender, where no web page or folder permission would suffice.

Apple is changing the permissions such software relies on at the same moment. PPC Land reported earlier today that macOS 27.3 beta 2 splits Apple's Device Control and Data Access setting and narrows Full Disk Access to chosen folders, changes that touch computer use agents which operate apps by reading the screen and clicking. Google's post does not say which macOS permissions computer use requests, or whether the feature has been tested against the 27.3 beta.

Why it matters for marketers

The obvious audience is the person with a tedious task. The less obvious one is everyone on the receiving end of the agent's clicks.

Start with reporting. Google's own example of pulling "metrics and screenshots from live dashboards across multiple regions" into Google Slides describes work that, at many agencies, sits with junior account staff. Dashboards in ad platforms and analytics tools sit behind logins, and according to 9to5Google, Gemini pauses at logins rather than handling them itself. The feature does not remove the human from the loop. It moves that person from assembling the deck to supervising its assembly.

Marketplaces come next. The resale example ends with Gemini preparing to "stage listings on resale platforms." Platforms have their own rules on automation. PPC Land reported on November 13 that eBay's user agreement bars buy-for-me agents and LLM-driven bots from its services without eBay's express permission, and that Meta's Muse agent had been blocked from eBay listings, according to The Verge. Google's word "stage" suggests the final step of publishing is left to the user, though the post does not say so explicitly.

Then there is traffic. HUMAN Security's October data, covered by PPC Land on November 13, put Google's agent at 0.8% of the AI agent traffic reaching the sites it protects, against 59% for Meta's Muse. Those figures predate today's feature. The Gemini app itself passed 1 billion monthly users according to Alphabet's October results, though Google does not disclose how many people use the Mac app.

Identification is unresolved. Two days ago, Google published a guide listing six reasons sites still block its agent after allowing it, built around the Google-Agent user agent and a published list of Google IP ranges. Will computer use on a Mac arrive at a website as Google-Agent, with signed requests from Google's address pool, or as an ordinary browser on a home broadband line? Today's post does not say. For publishers counting sessions and advertisers paying for clicks, the answer determines whether an agent's visit can be told apart from a person's.

What the post does not say

Google's announcement does not name the Gemini model behind computer use, say where the model processing takes place, list the macOS permissions involved, or state whether screenshots from the background window are retained or used for training. It does not define "key actions," name any blocked categories of apps or websites, describe defences against prompt injection from pages or files the agent reads, or give a date for other countries, languages, Windows or free accounts. It also gives no usage limits per plan.

Key facts

  • Feature: Computer use in the Gemini app for Mac
  • Announcement: "Let Gemini tackle repetitive chores on your Mac" (browser title "Automate Mac tasks with Gemini computer use"), Google blog, by Michael Friedman, Group Product Manager, Gemini App
  • Published: November 19, 2026, 17:00 UTC
  • Source URL: https://blog.google/innovation-and-ai/products/gemini-app/computer-use-mac/
  • Availability: "Computer use is rolling out in English in the U.S. in the Gemini app for Mac."
  • Footnote 1: "Requires an Apple Silicon Mac running macOS Sequoia 15.0+." "Available for personal Google accounts 18+, available to Google AI Plus, Pro and Ultra users."
  • Invocation: "@Computer use" in the prompt (shown in Google's hero image; described by 9to5Google)
  • Download: gemini.google/mac
  • Plan prices (as reported by PPC Land on October 4, 2026): AI Plus $4.99 a month, AI Pro $19.99, AI Ultra $99.99 or $199.99
  • Related Google help pages: "Use the Gemini app on Mac" (support.google.com/gemini/answer/17011627); "Manage your tasks & workflows in Gemini Apps" (support.google.com/gemini/answer/17094507); "Assign tasks in the Gemini app on Mac" (support.google.com/gemini/answer/17208717)

Timeline

Summary

Who: Google, through an announcement by Michael Friedman, Group Product Manager for the Gemini app. The feature reaches US users of the Gemini app for Mac who hold personal Google accounts, are 18 or older and subscribe to Google AI Plus, Pro or Ultra.

What: Computer use, a capability that lets Gemini operate desktop applications, local files and websites through their interfaces, invoked with "@Computer use" and run in an isolated background window with a live preview in the chat. Users choose which apps Gemini can access, can pause it at any time and approve key actions before they happen.

When: Today, in a post published at 17:00 UTC, two days after Google made its task mode free in six countries and 142 days after Gemini Spark first reached the Mac.

Where: In the Gemini app for Mac, on Apple Silicon Macs running macOS Sequoia 15.0 or later, in English and in the United States only at launch.

Why: Google presents the feature as a way to remove repetitive, multi-click work, from batch photo edits and shipping forms to dashboard-based slide decks and resale listings. For the marketing community, it adds a screen-driving agent from Google, at a starting price of $4.99 a month, while leaving open how its activity on websites will be identified.