Google shipped computer use for its Gemini app on Mac on Thursday, letting the AI drive desktop apps, local files, and websites on a user’s behalf — roughly a year after rivals like OpenAI, Anthropic, and Meta put agentic control in their own desktop applications.
Other leading AI model makers, like OpenAI, Anthropic, and Meta, have offered computer-use features in their desktop apps for some time, while Google has spent the week rolling out adjacent pieces — the broader release of Gemini’s in-app “Tasks” agent and early access to “CC,” a family-organization agent announced in September.
What Google is shipping today is narrower than the marketing suggests, and the fine print is where this story actually lives. The feature works only in English, only in the US, only for paid subscribers — Google AI Plus, Pro, or Ultra — and only on Apple Silicon Macs running macOS Sequoia 15.0 or newer, BigGo Finance reported. Google gave no timeline for other markets or platforms.
With computer use in the Gemini app for Mac, you can now hand off tasks across your folders, apps, and the web, while your screen and cursor stay yours. Watch Gemini click, scroll, and type in a live background window, turning a quick sketch into a rigged 3D model in Blender.
“feature to the macOS desktop app for its conversational AI” Google announced on November 19, 2026 that it has added a “Computer Use” feature to the macOS desktop app for its conversational AI “Gemini,” enabling the AI to operate the user's computer on their behalf.
Here’s what it does. A user describes a multistep task; Gemini runs it inside an isolated background window with a live preview in the chat, leaving the screen and cursor free.
Users can pause, stop, and approve actions before they happen, and choose which apps the agent can touch — safeguards Google spells out in its own announcement. Invoke it with @Computer use in a prompt, per Android Authority’s writeup of the rollout.

Two companies, two definitions of “agent”
The contrast Google itself drew this week is telling. Tasks, the renamed Gemini Spark which has been in beta since May running on the Gemini 4 Hydrogen model, handles functions such as email search, calendar entries, shopping, and phone reservations inside the app, and Google is making it available for free in six countries, including Japan.
Computer use, by contrast, operates outside the app, on the machine itself, and it’s paywalled from day one. Two agents, two trust models, two price tags.
Google’s demo, as Android Authority described it, is the flagship use case: a parent asks Gemini to turn a child’s butterfly sketch into an animated 3D model in Blender, then bring it to life in an augmented reality web app. Batch photo edits, shipping quotes from PDFs, presentations assembled from data across multiple websites — the pitch is that dozens of clicks become one prompt.
The outlet’s verdict: the convenience comes with security risks serious enough that handing over control is, in its view, a bad idea. That’s the sharpest divide in the coverage — Google presenting a live-preview showcase, AppleInsider arguing you shouldn’t use the thing at all.
Why the permissions stack should worry you
The mechanics explain the concern. Before Gemini can operate apps, users must enable computer use and grant macOS Accessibility plus Screen & System Audio Recording permissions, AppleInsider reported. Gemini then requests permission per app — and “Always allow” extends access beyond a single task, though Google says sensitive actions still need extra confirmation.

The dangerous part is what those permissions touch: apps already signed in to personal or professional accounts. Per Google’s own support documentation, cited by AppleInsider, Gemini can access active sessions in native Mac apps and even read multifactor authentication codes visible on the desktop or in messaging apps.
An AI agent with a window into your MFA codes is not a chatbot making a wrong suggestion — it’s a wrong action with account-level consequences.
Google does build in some containment. Backups generated during agent sessions expire after 24 hours or when the user submits another prompt, whichever comes first — AppleInsider’s reporting, and its note that these are documented risks, not evidence of routine damage.
The Gemini app is now on Mac. With this new desktop app, you can access Gemini from any screen with Option + Space and share your window to get answers based on the documents, code, or data you're working on.
Apple is moving the goalposts this same week
Timing is the other half of the story. Apple introduced new permission controls in the macOS 27.3 developer beta 2 on November 16 — three days before Google’s announcement — letting users restrict which apps an automation tool can control or which folders an app may access, AppleInsider reported.
The Verge adds the framing that Gemini is launching just ahead of Apple’s new restrictions on apps seeking full disk access, restrictions driven by exactly the risks agentic computer use creates. Google’s agent lands on a platform whose owner is actively tightening the doors it walks through.
The late-arrival economics
Google released the Gemini 4 Neon model on November 18, positioned between the top-tier Carbon and the speed-focused Hydrogen, cutting per-request processing cost by 32% versus Carbon under standard reasoning settings and scoring 61 on the Artificial Analysis Intelligence Index at maximum reasoning. Computer use rides on a model lineup Google is actively repricing — the paid tiers are the gate, and the agent is the reason to pay.

| Requirement | Detail |
|---|---|
| Launch date | November 19, 2026 (US) |
| Hardware | Apple Silicon Mac |
| OS | macOS Sequoia 15.0 or later |
| Plans | Google AI Plus, Pro, Ultra |
| Region / language | US, English only |
| Age / account | 18+, personal Google account |
Where the agentic race actually stands
Strip the buzzwords and the competitive picture is simple: OpenAI, Anthropic, and Meta defined the computer-use category in desktop apps; Google is following with a tighter geographic and linguistic footprint, a subscription wall, and a host platform pushing new restrictions simultaneously. The Verge’s reporting frames it as Gemini arriving ahead of Apple’s crackdown; AppleInsider frames it as a feature users should decline. Both can be true.
Google, the company behind Gemini, was founded in 1998 by American computer scientists Larry Page and Sergey Brin, and went public in 2004.
It has since grown from its original search engine into one of the world’s most valuable brands — Google Search and YouTube are the two most-visited websites worldwide — and in 2015 it was reorganized as a subsidiary of parent company Alphabet Inc., where Sundar Pichai, Google’s chief executive since 2015, also took the top job in 2019.
The Gemini push sits inside that broader AI portfolio: the company already fields AI chips known as TPUs, the TensorFlow machine-learning framework, and the DeepMind research arm, and its Search product began presenting AI Overviews in 2024. Advertising still dominates the business, accounting for 73% of Google’s revenues in 2025 — context worth keeping in mind as the company experiments with subscription-gated features like computer use.
For now, the sensible read sits in AppleInsider’s own closing advice: use Gemini to plan and prepare tasks without granting permission for potentially irreversible changes. The agent works in an isolated window, shows its steps live, and can be stopped at any point — but the permissions it asks for, and what they expose, are the real product decision. Google hasn’t said when any of this leaves the US, or English.