Google is upgrading the Gemini app for Mac with computer use capabilities, letting the AI click and type on a user's behalf across local desktop applications and the web. The feature began rolling out Thursday in US English for Google AI Plus, Pro, and Ultra subscribers, according to 9to5Google, with Bloomberg and TechCrunch also reporting the launch and Google itself announcing it on the official blog under the banner "Let Gemini tackle repetitive chores on your Mac."
Until now, the Task functionality in the Mac app — previously known as Gemini Spark — could only work with local files in combination with Connected Apps like Gmail or Google Calendar. Computer use extends that reach considerably: Gemini can now navigate menus, fill out forms, and complete multi-step tasks inside desktop applications and on websites, operating in what Google describes as an isolated background window with a live preview right in the chat. Your main screen and cursor stay free while the agent works. For a sense of how aggressively the major labs are shipping agentic features this month, see our roundup of the latest AI developments.
What Gemini can actually do on a Mac
Google's pitch is that most everyday work spans applications that don't talk to each other, and that an agent working in the background can absorb the tedious middle steps. The company and its press coverage highlighted several concrete use cases:
- Speeding up creative workflows — batch color-grading photos in design software or setting up complex 3D scenes.
- Skipping manual form entry — extracting weights, dimensions, and addresses from local PDFs to fill out multi-box shipping quotes.
- Turning live data into presentations — pulling metrics and screenshots from dashboards across regions and assembling them into a Google Slides deck.
- Batch-listing items for sale — analyzing clothing photos from a local folder, researching pricing, logging item details, and staging listings on resale platforms.
Users invoke the capability by adding @Computer use to a prompt. In the app's Customize > Connectors settings, new "Desktop" integrations for Blender and Autodesk Fusion appear alongside the existing Git and Obsidian connectors — a notable nod to creative and engineering professionals, since those applications are central to 3D design and mechanical workflows.
The safety guardrails
Handing an AI control of your keyboard and mouse is a trust exercise, and Google has built in several limits. Users can stop Gemini at any time with one click, and choose which applications the agent is allowed to touch. Gemini pauses for logins and payment steps, and sensitive actions require explicit confirmation before proceeding. The isolated-window design adds another layer: the agent works in its own virtual space rather than hijacking the user's active desktop, with a live preview showing what it is doing.
Those guardrails echo the approach Google has taken with agentic features elsewhere in its stack — the company recently upgraded Gemini's agentic Task mode more broadly, and added incognito mode to AI search as part of the same privacy push. They also land amid rising scrutiny of what AI agents can do on users' behalf: security researchers have spent much of 2026 documenting prompt-injection attacks against coding agents and browser-based agents alike, making the pause-for-payments and confirmation requirements more than ceremonial.
Another salvo in the agent platform war
Computer use on Mac extends Google's agent ambitions into territory that rivals have been circling all year. Anthropic pioneered the "computer use" category when it opened Claude's ability to operate a desktop in 2024, and OpenAI has since pushed agent features deep into ChatGPT. Microsoft, meanwhile, has leaned on its Windows installed base to ship agent defenses and Copilot-driven automation at the OS level.
What distinguishes Google's entry is distribution and framing. Gemini for Mac ships to a consumer audience rather than developers, runs locally on the machine where the work happens, and is positioned around chores — the repetitive, multi-app tasks that people procrastinate on — rather than around autonomous work. The subscriber gating is telling, too: AI Plus, Pro, and Ultra are Google's paid tiers, making computer use another perk in the subscription bundle as the company converts Gemini's billion-plus user base into paying customers. Google also recently made Gemini Tasks free for more than a billion users, suggesting a ladder in which entry-level agentic features build the habit and premium tiers like computer use monetize it.
The rollout starts in US English today, and — as with most Gemini features — broader language and regional availability will likely follow. Mac users on free tiers will need to wait, or upgrade, before handing their paperwork to Google's agent.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →