Google gave Gemini hands yesterday, at least on the Mac. On 19 November, Google announced computer use in the Gemini app for Mac: describe a multi-step job in plain words, and Gemini clicks, types, navigates menus and fills in forms across your local apps and the web, working in an isolated background window while your own screen and cursor stay free. I have been waiting for this kind of feature to reach ordinary users, because the gap between a chatbot that answers and an assistant that acts is exactly the gap that eats a freelancer's day. This is a real step across it, and it comes with two catches worth knowing before you get excited.
What it actually does
You trigger it by adding "@Computer use" to your prompt, then describing the job. Gemini works in an isolated background window, and you can watch every step in a live preview without handing over your own screen or cursor. It does not run the job itself for you: it performs the same actions you would, but on your behalf, and it asks you to step in when it reaches a login, a payment or any other sensitive step. Here are five jobs it handles:
- Batch colour-grade photos in your design software, or set up complex 3D scenes.
- Skip manual form entry: pull weights, dimensions and addresses out of local PDFs and fill in multi-box shipping quotes.
- Turn live data into ready-to-present decks: grab metrics and screenshots from live dashboards and build a presentation in Google Slides.
- Batch-list items for sale: analyse clothing photos in a local folder, research pricing, log the details, and stage listings on resale platforms.
- Organise files into folders, and repeat the same task across a batch of files.
The safety controls are the part I read twice. You choose which apps Gemini can access, you can watch every step live, pause at any time, and approve key actions before they happen. Google says it will pause for logins and payments, with confirmation needed for sensitive actions. That matters, because an assistant that can click can also delete files or share them, and the difference between useful and dangerous here is entirely in those controls.
The angle nobody is spelling out: this automates the boring half of freelance work
For most freelancers, the paid work is not the hard part. It is the quoting, the filing, the renaming, the listing, the copying numbers between tabs. That is exactly the work computer use targets: the jobs you already know how to do, repeated often enough to be boring, but too fiddly to hand to a template. If Gemini can do a shipping quote from a PDF while you finish the real work, the Mac just became a two-person office — and that is a bigger shift for a solo freelancer than any new model benchmark.
The same logic applies to the jobs I have written about this week. In my guide to switching to Microsoft MAI Code 2 in GitHub Copilot, the point was choosing the right model for the right job; here the point is choosing which jobs are worth automating at all. And in my piece on Meta's Muse, I argued the agent era is arriving as a price war — Google's move is the same war, fought on your desktop instead of in a model leaderboard.
The two catches
First, the rollout is narrow: English only, US only, adults only, personal Google accounts only, and a paid plan is required — Google AI Plus, Pro or Ultra, with plans starting at $4.99 a month. Second, your hardware is a gate too: you need an Apple Silicon Mac running macOS Sequoia 15 or later. Most of us, in other words, have to wait.
Trust is the bigger question, and Google has answered it with a design choice I like: Gemini pauses for logins and payments and asks for confirmation before sensitive actions, and the isolated background window means it is not driving your cursor while you work. That is the right shape for this kind of tool — and it is also why Apple's reported move matters. If Apple builds its own agent and tightens the controls on third-party agents, the question for Mac users will not be whether an agent can run their computer, but whose agent they trust to do it. I wrote about Apple's plan in detail this week, in my piece on Apple building its own AI agent for Mac while Gemini got there first.
Bottom line
Gemini's computer use is real, useful and narrow: five clear jobs, strong safety controls, and a rollout that excludes most of us for now. If you are in the US, on an Apple Silicon Mac, and paying for a Google AI plan, try it on the most boring job in your week — a shipping quote, a batch of listings, a folder of files — and watch it work in the live preview. For everyone else, the wait is the story: the first mainstream computer-use agent has shipped, and the race to bring it to the rest of us has just started.
Sources: Google, Automate Mac tasks with Gemini computer use, by Michael Friedman, 19 November 2026, blog.google.
Originally published on Ai Game Changer Pro


