Gemini operates Mac apps: background computer use starts in the US
The shift is from answering questions to acting on screen: on 19 November Google announced computer use in the Gemini app for Mac, enabling clicking, typing and form filling across local files, applications and websites.
This is a two-day-old announcement not previously covered in this column, rather than a launch taking place today.
PPC Land reported the arrival on 19 November, distinguishing interface control from the agent’s other capabilities.
## A separate window does not mean an unsupervised computer
Google describes an isolated background window with a live preview in the conversation, leaving the user’s main screen and cursor free.
The company lists app-access choices, the ability to pause work and approval of key actions before they happen.
The important wording is “key actions”: we do not extend it into a promise that every individual click requires separate consent.
An isolated window alone also does not establish on-device inference or guarantee that no data leave the Mac; no such guarantee is made here.
## The rollout does not automatically include Italy
The announced availability is in English in the United States.
The official footnote requires an Apple Silicon Mac running macOS Sequoia 15.0 or later, a personal Google account for someone aged 18 or over, and access through Google AI Plus, Pro or Ultra.
These are rollout conditions: the feature is not being presented as available to every Mac, work account or Italian user.
## Use cases are the vendor’s examples
Google’s examples include extracting weights, dimensions and addresses from local PDFs to fill shipping quotes, and building presentations from dashboard metrics and screenshots.
These illustrate proposed workflows rather than the results of our own independent tests of accuracy or time saved.
The editorial point is a change in responsibility: when an assistant acts inside applications, permissions and outcome review become part of the job, not a minor interface detail.
For work involving documents and forms, a cautious approach is to check what is read, what is written and which action has actually been completed.
