Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarThe Decoder→ original

Gemini for Mac can now run tasks in the background while you work

Google has added a Computer Use mode to the Gemini app for Mac. The AI clicks, types and navigates menus on its own across apps, files and websites. You describe a goal, for example filling in a shipping form using data from a PDF, and Gemini works in an isolated background window without taking over your cursor. Every step is visible in the chat, and you can hit pause at any moment. For product teams, it is a template for making desktop agents controllable.

AI-processed from The Decoder; edited by Hamidun News
Gemini for Mac can now run tasks in the background while you work
Source: The Decoder. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceThe Decoder ↗
Editor-in-Chief
Key metric #1Currently available only in English and only in the US
Key metric #2The user chooses which apps Gemini can access
Google's examplesbatch editing of photos and 3D scenes, building presentations

Picture an ordinary working evening: you need to sort through forty photos, fill in a shipping form using data from a PDF, and put together a draft presentation. It is half an hour to an hour of routine that pulls your attention away from real work. Google proposes handing it off to Gemini. The Mac app now has a Computer Use mode, and it handles that routine in the background while you do something else.

What exactly did Google show?

Computer Use performs multi-step tasks in three places at once: local files, desktop apps and websites. The user states a goal in plain language. From there, Gemini clicks, types and works through menus and forms the way a person at a computer would.

The key engineering detail is where all of this happens. The agent works in an isolated background window, and the chat shows a live preview of its actions. As a result, your screen and cursor stay yours: you can write an email or edit code while Gemini does its part. Agents of this type have usually grabbed control of the machine, which made working alongside them awkward.

The examples in Google's announcement are fairly down-to-earth:

  • batch editing of photos and 3D scenes;
  • filling in shipping forms using data from local PDFs;
  • building presentations.

How does control work?

For an agent that presses buttons on your computer, trust matters more than polished demos. Google has built several layers of control. The user chooses which apps Gemini can access at all. The user sees every step in real time and can pause the task at any moment. Before key actions, the agent asks for confirmation in advance.

It is a sensible design: the agent's freedom is limited to a list of approved programs, and irreversible steps such as submitting a form go through a human. For anyone designing desktop or browser agents, it is a ready-made interface pattern: isolation, transparency and confirmation points instead of a blind "just do everything yourself."

What is still unknown

Now the honest part: the announcement contains no numbers. Google gives no data on how often the agent completes a task successfully, how many steps it can sustain, or how it behaves in non-standard interfaces. All the examples come from the company's own materials, and there are no independent tests yet. Google also does not say which apps are supported best.

There is a geographic limitation as well. Computer Use is rolling out today only in English and only in the US. The Gemini app for Mac can be downloaded, but the mode itself is not available everywhere. Users in Russia and Europe will have to wait or look for workarounds.

Competition and takeaways for practitioners

The Decoder notes that OpenAI is also actively working on Computer Use and, in the outlet's assessment, is pulling noticeably ahead with its Astra model. The piece gives no details of this comparison, so treat it as the outlet's opinion, not a measured result. The direction, though, is clear: major labs are betting on agents that work through the ordinary graphical interface, not only through APIs.

What to do with this. Product teams should look closely at the background-window pattern: it removes the main annoyance of agents that get in the way of the person working. Developers whose apps need to coexist with such agents should check how clear and predictable their interfaces are: forms without explicit labels and non-standard buttons will break automation. For founders building niche desktop agents, this is a signal: the basic scenario of "describe the goal and I'll click for you" is now part of a standard Google product, and competing will mean depth in a specific domain, not the mere ability to control a computer.

Story

Key Questions Answered

What exactly did Google show?

Computer Use performs multi-step tasks in three places at once: local files, desktop apps and websites. The user states a goal in plain language. From there, Gemini clicks, types and works through menus and forms the way a person at a computer would. The key engineering detail is where all of this happens. The agent works in an isolated background window, and the chat shows a live preview of its actions. As a result, your screen and cursor stay yours: you can write an email or edit code while Gemini does its part.

How does control work?

For an agent that presses buttons on your computer, trust matters more than polished demos. Google has built several layers of control. The user chooses which apps Gemini can access at all. The user sees every step in real time and can pause the task at any moment. Before key actions, the agent asks for confirmation in advance. It is a sensible design: the agent's freedom is limited to a list of approved programs, and irreversible steps such as submitting a form go through a human.

What is still unknown?

Now the honest part: the announcement contains no numbers. Google gives no data on how often the agent completes a task successfully, how many steps it can sustain, or how it behaves in non-standard interfaces. All the examples come from the company's own materials, and there are no independent tests yet. Google also does not say which apps are supported best. There is a geographic limitation as well. Computer Use is rolling out today only in English and only in the US.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…