Google Brings Computer Use to Gemini on Mac, Months Behind Rivals
Google's Gemini desktop app for Mac can now control the computer to complete multi-step tasks, launching in English in the U.S. behind OpenAI, Anthropic and Meta's rival agents.

- Google announced computer use in the Gemini Mac app on Thursday, letting the AI operate native apps and the web on the user's behalf
- The feature is English-only, U.S.-only at launch and requires a Google AI Plus, Pro, or Ultra subscription
- OpenAI, Anthropic and Meta already offer agentic features in their desktop applications, putting Google behind
- Gemini's 'Tasks' agent feature rolled out broadly only this week; the 'CC' family agent from September just began admitting early testers
- Users can pause, stop and approve the agent's actions; no expansion timeline was given
Google announced on Thursday that its Gemini desktop app for macOS can now control a user's computer, completing multi-step tasks across native Mac applications, local files and websites on the user's behalf.
The launch puts Google behind its competitors. OpenAI, Anthropic and Meta have already shipped agentic features in their desktop applications. Google only this week announced the broader rollout of Gemini's "Tasks," an in-app feature that lets the AI act more as an agent — signing into websites, making online purchases, or booking appointments. Google also only recently began admitting early testers to its revamped "CC" AI agent, announced in September, which is designed to help families stay organized across email, calendars, chats and tasks.
What can the Gemini agent actually do?
Computer use is a distinct category of agentic capability. The AI takes direct action on the user's machine rather than merely answering questions. In Gemini's case, that machine is a Mac.
The feature handles multi-step workflows that span local files, apps and the web. After the user instructs the AI on what to do, a live preview of the agent's work appears inside the chat. The user's screen and cursor stay free, so other work can continue in parallel.
Google suggests several concrete use cases:
- Filling out forms
- Repeating the same task across a batch of files
- Organizing files into folders based on specific criteria
- Creating slideshows from files or information in online dashboards
As with other AI agents offering computer use, users can pause, stop, and approve Gemini's actions before they happen. That approval layer matters, because an agent with cursor-level control of a Mac can touch sensitive data, and Google's announcement did not detail how failures or errors are handled mid-task.
The honest read: the capability matches what rivals already offer, and the safeguard model mirrors the industry standard rather than exceeding it.
Who gets access, and what does it cost?
Availability is narrow at launch. The feature works in English, in the U.S. only, and requires two things: the Gemini app for Mac and a Google AI Plus, Pro, or Ultra subscription. Google did not say when the feature would expand to other markets or platforms.
That silence leaves open the obvious questions. No date for other languages. No date for Windows. No stated plan for a free tier. Each of those gaps will determine whether computer use becomes a mainstream Gemini capability or stays a subscription perk for a small slice of U.S. users.
Why the timing matters
The pattern across Google's recent announcements is one of catch-up. Tasks rolled out broadly only this week. The CC family agent, announced in September, only just started admitting early testers. Computer use on Mac arrives after OpenAI, Anthropic and Meta already offer comparable agentic features in desktop apps.
Google is shipping agents, but it is shipping them later than its competitors and into a narrower market. The live-preview design and the pause-stop-approve controls are sensible, and the batch-file and slideshow use cases target genuine tedium — yet nothing in Thursday's announcement describes a capability the competition lacks.
What to monitor next: whether Google commits to a timeline for international and Windows expansion, and whether early U.S. users report reliable completion of multi-step tasks once the agent moves beyond controlled demos.
via techcrunch.com (Source)
More from Elena Vasquez
Show full bio
Correspondent covering people, launches and day-to-day practice in Search rankings and algorithm updates, generative engine optimization at SERP Journal.
227 articles