Dek: Google's Mac app can now take a multistep job and finish it in apps it was never connected to, inside a window you don't have to watch. The part worth reading is where it stops and asks.
What Gemini computer use does when you type @Computer use
The mechanic is small and easy to miss. In a Gemini chat on macOS, you add @Computer use to the end of a prompt, and the app takes it from there.
It opens a separate window and works inside it. Your own screen stays yours. So does your cursor.
That sounds like a detail until you've watched an automation tool grab the pointer mid-sentence while you were trying to answer an email. Working in a second window is the difference between a tool and a houseguest.
Google announced the feature on Thursday 19 November. TechCrunch and 9to5Google both covered it the same day, and the shape of the announcement is unusual for this company: no model name, no benchmark, no claim about getting there first.
Inside that second window, Gemini clicks menus, fills forms and types into fields. The apps it drives include software that was never wired into Gemini through a connector. That's the break from the older Task feature in the Mac app, which could reach local files only through Connected Apps such as Gmail and Google Calendar.
The Gemini for Mac desktop agent reaches apps nobody connected
Scope is where this gets interesting for anyone with a messy desktop.
SiliconANGLE reports Gemini can handle tasks made up of dozens of actions. It can also take the output of one job and use it as the input for the next, so a visual asset built in a design tool can become the starting point for a small app in Visual Studio Code.
Chaining is the part that changes a work day. One step is a party trick. A chain of five is an afternoon you get back.
Google also added Desktop entries under Customize > Connectors. Blender and Autodesk Fusion appear there, next to the Git and Obsidian integrations that were already in the app. Those two names tell you who Google expects to use this first: people with 3D scenes and repositories open.
Four jobs Google says Gemini can run unattended
Google's own examples are specific, which makes them easy to judge.
In a design app, Gemini can batch color-grade a folder of photos or set up a complex 3D scene. For shipping, it can pull weights, dimensions and addresses out of local PDFs and fill in a multi-box quote. For reporting, it can collect metrics and screenshots from live dashboards across regions and assemble them into a Google Slides deck. For resale, it can sort through a folder of clothing photos, research pricing, and stage the listings.
Only the shipping example is boring in the useful way. Pulling numbers out of PDFs and into a form is the task most small offices still do by hand, and it's the one where a wrong digit costs real money.
Permissions, previews and the logins Gemini won't touch
Controls matter more here than in a chat window, because the thing clicking around has access to your files.
You choose which apps Gemini can reach. You can watch every step in a preview pane inside the chat, pause mid-run, and approve actions before they happen. Michael Friedman, a group product manager for the Gemini app, described the intent in the blog post: "Privacy and safety are built in: you choose which apps Gemini can access, watch every step live, pause at any time, and approve key actions before they happen."
Google says the desktop agent will stop for logins and payments, and that sensitive actions need a confirmation before they go through. That's the right place to draw the line. A password field is where an automation mistake turns into an account problem.
Google AI Plus, Pro and Ultra, in US English only
Availability is the shortest section because Google said the least here.
Computer use rolled out on 19 November in US English, for Google AI Plus, Pro and Ultra subscribers. That's the whole list. No Windows date, no EU date, no word on when the feature reaches other languages.
For a US subscriber on one of those tiers, the feature is in the Mac app now. For everyone else, this is a launch you read about and wait on.
Not yet, anyway.
When does it reach Windows or the EU? Google didn't say, and the blog post doesn't either.
The gate also says something about cost. A job that clicks through dozens of steps burns compute in a way a text reply doesn't, and Google is charging it to the tiers that already pay.
Muse for Mac, OpenAI's Dots and the feature Google skipped
Google isn't early. TechCrunch notes that OpenAI, Anthropic and Meta shipped comparable capability first, and that Google only broadened the rollout of Gemini Tasks on 17 November, two days before this announcement.
Meta's Muse for the Mac also drives local apps. The difference is a cloud virtual machine that keeps working when your Mac is off or asleep. OpenAI's Dots does something similar, according to SiliconANGLE.
That gap is the honest headline for a laptop owner. A desktop agent that needs the lid open is a faster pair of hands. One that keeps going after you close the lid is closer to staff.
What Gemini for Mac still hasn't answered
Three gaps are worth holding Google to.
The first is reliability. Nobody outside Google has measured how a 40-step job behaves when step 31 hits a dialog box that wasn't in the plan. The preview pane lets you watch, but watching doesn't tell you the odds.
The second is the model. Google won't name it, which makes cost, speed and failure hard to reason about when a job goes sideways.
The third is the rest of the world. A Mac-only, US-only, English-only launch leaves most of Google's user base reading release notes about a feature they can't open.
If you're on AI Plus or above in the US, start with a job that has a clear finish line, like the PDF-to-shipping-quote example. Watch the preview for the first few runs and keep the app list short. Trust in a desktop agent gets built one finished task at a time, and Google hasn't yet shown what happens when the task doesn't finish.






