Gemini Can Now Touch Your Mac's Screen — 'Computer Use' Starts in the US, on Paid Plans
The result first
Gemini has gained “hands” — but they do not reach this house yet.
- On November 19 (US time), Google added “computer use” to the Gemini app for Mac: describe a job in words, and Gemini clicks, types, works menus and fills in forms on your behalf, across files, desktop apps and websites
- The work runs in a separate background window, with a live preview in the chat; you choose which apps Gemini may operate, you can pause or stop at any time, and important actions — logins, payments — stop and ask for your approval first
- The honest limits are three: it rolls out from the US in English (no date announced for Japan); it is for paid plan subscribers (Google AI Plus, Pro, Ultra), not free; and it is Mac only — Apple silicon, macOS Sequoia 15.0 or later, a personal Google account of someone 18 or older
- Our household’s centre is talking with Moka on an Android phone. We have neither a Mac nor a paid plan here, so this is a feature we cannot try yet — and this article says so plainly
Hello, I am Kurashiki Bouzu (kurashiki-bouzu), born in 1971. I am a former funeral director, and I now live in an old house in Kurashiki, Okayama, writing articles every day in conversation with my AI butler, Moka.
Word is that Gemini can now touch a Mac’s screen directly — doing jobs that span apps and files, in a window behind yours. But this time, too, the honest lines matter. It starts in the US, and it is a paid-plan story. Here it is, in order.
What was announced — Gemini gained “hands”
Google announced on its official blog on November 19 (US time) that it had added a feature called “computer use” to the Gemini app for Mac.
Tell it the job you want done, in words, and Gemini performs the clicks, typing, menu operations and form entries for you. It works across files on the Mac, desktop apps and websites, and can handle multi-step jobs that span several apps, Google says.
The four examples Google gives are:
- Adjusting the colours of a batch of photos in design software
- Pulling weights, dimensions and addresses out of a PDF and entering them into a shipping-quote form
- Gathering numbers and screenshots from dashboards for several regions and turning them into a Google Slides deck
- Analysing photos of clothes in a folder, researching prices, and preparing listings for a resale service
The way in is to add “@Computer use” to your request, according to 9to5Google’s reporting.
Until now, Gemini’s “Tasks” worked inside connected apps or dedicated mechanisms. What is new this time is the step to operating the computer’s screen itself, in the person’s place. Features that let an AI operate a computer came earlier from OpenAI’s apps and Anthropic’s Claude; Google is the latecomer. That it arrived inside the ordinary Gemini app is, in my view, this announcement’s distinguishing point.
The safety design — a back window, and approval first
When you let something touch your screen, safety is the worry. In Google’s announcement, the design is:
- Gemini’s operations run inside a dedicated window in the background, separate from your screen
- The chat screen shows the work as it happens (a live preview), so your own screen and cursor stay free. You can keep typing or talking while Gemini works
- You choose yourself which apps Gemini is allowed to operate
- You can watch the steps in real time, and pause or stop at any time
- Important actions ask for your approval before running. At logins and payments, it stops and asks, Google says
Not “leave it entirely to the machine,” but “let it proceed in the back window, and stop at the important points for a person’s approval.” This shape felt close to the line we draw in this house when work is entrusted to Moka — at the important steps, the owner’s confirmation is taken.
The honest lines — from the US, from paid plans, Mac only
Now the honest part. There are three conditions you cannot see from the announcement’s headline alone.
First: the rollout begins in the US, in English, and spreads from there. No date for Japan has been officially announced. “Unknown” is the honest answer right now.
Second: it is for subscribers of the paid plans — Google AI Plus, Pro and Ultra. It is not free. The original news article (TechnoEdge) did not state this condition. It came out by checking Google’s official blog against overseas media reporting.
Third: it is Mac only. The stated requirements are an Apple silicon Mac, macOS Sequoia 15.0 or later, and a personal Google account of someone 18 or older (carried by overseas media as a note in the official blog). No Windows version was announced this time.
The centre of our life here is talking with Moka on a smartphone (Android). A Mac and a paid plan are, for now, not part of this house. So this feature is “the announcement arrived, but we cannot try it here yet.” What we cannot try, we write down as cannot try.
In our project, this is a candidate for a “new kind of hand”
Since November 19, this house has been running a project called “Putting Gemini to work with Moka”: we ask Gemini to do real errands and record what it could and could not do. The candidate errands are divided between “task” errands like phone calls and shopping, and errands where Gemini works inside Google apps.
This “computer use” is a kind of errand that is in neither shelf yet. It does not work inside a connected app; it operates the computer’s screen itself. You could say Gemini has grown “hands,” and as a candidate errand for the project, it is a new step.
For example, errands that open an app’s screen and do the preparation, or repetitive work spanning several apps, would be worth trying with these “hands” once they can be used in Japan. For now, though: US, English, paid plans, Mac only. In the project’s planning, its position is “a candidate to try when it becomes usable in Japan,” kept in the record.
One difference from Moka is also visible. Moka (Muse) does not touch the screen directly; she advances her owner’s work using set tools and records. Gemini’s “computer use” clears errands by touching the screen. Neither is above the other; the shapes differ. Confirming that difference with the real thing is, I think, the interesting part of this project.
Money spent, and sources for the numbers
Honestly:
- The money we spent this time was 0 yen. We only read the announcement and reporting, checked them, and compared them with the project’s record
- The announcement’s contents (timing, design, safety, the four examples, rollout from the US in English) are from Google’s official blog (2026-11-19)
- That it is for paid plan subscribers (Google AI Plus, Pro, Ultra), and the “@Computer use” way of asking, were checked against 9to5Google and TechCrunch reporting
- The operating requirements (Apple silicon Mac, macOS Sequoia 15.0 or later, personal account of someone 18 or older) are from PYMNTS and PPC Land reporting, which carried the official blog’s note. They could not be confirmed in the official blog’s own text, so they are placed here as figures agreed by two outlets
- No date for Japan appears in the official blog or the reporting. That is why “unknown” is written
From here
This article is a record that “the announcement arrived.”
Whether “computer use” becomes usable in Japan, and whether it ever spreads to the free tier — when that day comes, we will ask it to do a real errand as a new candidate in the “Putting Gemini to work” project, and record what it could and could not do, as it happened.
Sources
- Google official blog, “Let Gemini tackle repetitive chores on your Mac” (2026-11-19): https://blog.google/innovation-and-ai/products/gemini-app/computer-use-mac/
- TechnoEdge, “Gemini operates your Mac for you: ‘computer use’ begins” (2026-11-20): https://www.techno-edge.net/article/2026/11/20/5778.html
- 9to5Google, “Gemini app for Mac adds computer use to control desktop apps” (2026-11-19): https://9to5google.com/2026/11/19/gemini-mac-computer-use/
- TechCrunch, “Google gives Gemini the ability to use your Mac for you” (2026-11-19): https://techcrunch.com/2026/11/19/google-gives-gemini-the-ability-to-use-your-mac-for-you/
- PYMNTS, “Google’s Gemini App Can Now Operate Apple Desktop Computers” (2026-11-19): https://www.pymnts.com/news/artificial-intelligence/2026/googles-gemini-app-can-now-operate-apple-desktop-computers/
- The plan and record of the project “Putting Gemini to work with Moka” (from 2026-11-19, this house’s own record)
- November 21, 2026 ― English draft (result-first rewrite of the Japanese article, not yet published)(6,946 characters)
- November 21, 2026 ― Published (all 7 languages at once, on the user's instruction). Japanese original: ledger No. 1383. New URL form with the ledger number at the end.(6,946 characters)
