The most interesting part of Google’s new Mac agent isn’t that it can click buttons. It’s that it does so without taking your screen away from you.
Google announced that computer use is now live in the Gemini app for Mac. You describe a multi-step task, and Gemini works across local files, desktop apps and the web, handling the clicking, typing, menus and forms. It runs inside an isolated background window, with a live preview in your chat. Your cursor and main display stay free, so you can keep taking calls or designing while it works.
That design choice matters. Anthropic’s Computer Use and OpenAI’s Operator both put an agent in front of a screen, and the usual complaint is that you either watch it or lose your machine for a while. A background window removes much of that friction. If it works reliably, it turns an agent from a demo into something you’d actually leave running.
Google’s examples show the intended range:
- Batch color-grading photos in design software or setting up 3D scenes.
- Pulling weights, dimensions and addresses from local PDFs to fill shipping quotes.
- Collecting metrics and screenshots from live dashboards to build a Google Slides deck.
- Analyzing clothing photos in a folder, researching prices and staging resale listings.
These are dull jobs, and that’s the point. The agent pitch works best on repetitive chores that cost dozens of clicks, not on open-ended reasoning where errors are costly.
Safety controls are part of the package. You choose which apps Gemini can reach, watch each step live, pause at any time, and approve key actions before they happen. That approval step is the right call. Agents that fill forms and post listings can do real damage when they misread a page, and every vendor in this category is still wrestling with prompt injection from web content.
The bigger picture is a race to own the desktop. Anthropic has pushed computer use for developers, Microsoft is building agents into Windows and Copilot, and Apple has been slow to ship anything comparable. Google getting a native Mac app with local file access puts Gemini on a platform where many developers and designers spend their day.
There are limits. The rollout is English only and U.S. only, and Google has shared no benchmarks or failure rates. Anyone can write a good demo about batch-listing resale items. Whether it holds up on messy real-world apps is the open question.
Still, this is worth trying if you’re on a Mac in the U.S. The background execution is a genuine usability gain, and it sets a bar that competitors will have to match. Download it at gemini.google/mac and test it on a task you hate doing.



