Pull to refresh
Logo
Google adds computer use to the Gemini app for Mac

Google adds computer use to the Gemini app for Mac

New Capabilities

A consumer AI agent gets mouse-and-keyboard control of desktop apps, three days after Apple tightened the rules for exactly that

Overview

Updated 42 minutes ago

Google's Gemini app for Mac can now click, type and scroll through desktop applications on your behalf. The feature, called Computer use, runs in a separate window with a live preview, and only touches apps you approve.

It arrived three days after Apple split macOS automation permissions in a beta of macOS 27.3, requiring fresh approval each time an app asks for control. Google has not said how Gemini will work under those rules. Its own help page lists the risks: Gemini inherits logged-in sessions inside native apps, text on screen can steer it into other approved apps, and permission for a code editor can mean shell access.

Why it matters

Your Mac now runs an AI agent inside logged-in apps, and the permission rules around it were rewritten three days earlier.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

1
Countries with access at launch
Computer use is rolling out in English in the United States only.
3
Google subscription tiers included
Google AI Plus, Pro and Ultra, on personal accounts.
18
Minimum account age
Personal Google accounts only; no work or school accounts.
3
Days between Apple's permission change and Google's launch
Apple split macOS automation permissions in the 27.3 beta days before the announcement.

Voices

Curated perspectives — historical figures and your fellow readers.

Sign up to generate historical perspectives on this story.

People Involved

Organizations Involved

Timeline

October 2025 November 2026

4 events Latest: Today
Tap a bar to jump to that date
  1. Gemini begins taking multi-step Mac tasks

    Today Rollout

    Users hand Gemini a task in a background window; it pauses for logins and payments and asks before sensitive actions.

  2. Google announces Computer use for the Mac app

    Product launch

    Rolling out in English in the U.S. to AI Plus, Pro and Ultra subscribers on Apple Silicon Macs running macOS Sequoia 15.0 or later.

  3. Apple splits macOS automation permissions

    Platform rule

    The macOS 27.3 beta 2 separates device control from data access and lets users approve each request instead of granting standing access.

  4. Computer use ships as a Gemini API

    Product

    Gemini 2.5 reads screenshots and returns click and type actions for developers to execute in their own environments.

Scenarios

1

Computer use spreads beyond U.S. English and the Mac

Likely Resolves by Q2 2027

Discussed by: Google's release notes and coverage in The Verge and Quartz, both of which note Google gave no timeline for other countries or operating systems

Google shipped the PC version before the Mac version, suggesting a staged rollout. The constraint is the permission layer, not the model: each operating system handles app control differently, and Apple's rules shifted days before launch. A second country or a Windows release would show Google can absorb that variation.

2

Apple's macOS 27.3 rules force a Gemini permission redesign

Likely Resolves by Apr 30, 2027

Discussed by: The New Stack and Decrypt, which both flag that Google has not explained how Gemini will work under the new controls

Apple's beta splits automation into two permission categories and lets users approve single requests rather than granting standing access. If that ships in the release build, an agent built around an always-allow grant may have to re-ask mid-task. Google can build against the new model or keep work inside a window it controls.

3

A documented prompt-injection incident on a Mac

Possible Resolves by Q2 2027

Discussed by: Google's own help page, which warns that untrusted on-screen content can push Gemini into other allowed apps

Computer use reads the screen, and the screen carries text written by other people: an email, a web page, a PDF. Google's documentation names this failure mode directly. One reported case of Gemini acting on hidden instructions would force either tighter confirmation prompts or narrower app permissions.

4

Workspace accounts get computer use

Possible Resolves by End of 2027

Discussed by: Google's help page, which limits the feature to personal accounts for now

The help page states the feature is not available for work or school accounts. The gap is administrative: a business needs a way to approve which apps an agent may open and to see what it did afterward. Until those controls exist, the feature stays consumer-only.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

March 1999

The Melissa macro virus (1999)

A Word document posted to an internet newsgroup carried a macro that emailed itself to the first 50 addresses in the victim's Outlook contacts. It spread faster than corporate mail filters could be updated, and some companies shut down mail servers to stop it.

Then

Antivirus vendors shipped signature updates within days, and enterprises tightened attachment policies.

Now

Microsoft spent the next decade adding warnings, then made macros opt-in for documents arriving from outside trusted locations. The default flipped from run to block.

Why this matters now

A document that executes steps on your behalf was the 1999 version of an app that clicks on your behalf. The fix was default-deny, which is what Apple's new permission prompts are reaching for.

April 2021

Apple's App Tracking Transparency (2021)

Apple required iPhone apps to ask permission before tracking users across other companies' apps and websites. Most users declined. Meta said publicly the change would cost it billions in advertising revenue.

Then

Opt-in rates landed far below what advertisers wanted, and measurement and attribution tooling was rebuilt.

Now

The permission prompt became the template platforms use to gate access to device data, including automation.

Why this matters now

Apple is running the same play on automation: split the permission, ask per request, show an indicator when one app drives another. Google shipped into that framework.

2016-2021

The robotic process automation boom (2016-2021)

Companies hired software bots to move data between enterprise applications, clicking through interfaces that had no programming hooks. Vendors including UiPath, Automation Anywhere and Blue Prism built businesses on it. The bots followed fixed scripts and broke whenever a screen changed.

Then

RPA cut the cost of repetitive back-office work but needed constant maintenance as interfaces shifted.

Now

The category established that clicking through a screen is a legitimate way to integrate software, and created the market model-driven agents now target.

Why this matters now

Gemini drops the fixed script. It reads screenshots and decides what to click, so it generalizes across apps and can also improvise in ways a script never would.

Sources

(10)