Google is developing autonomous control features for its Gemini macOS app, enabling the AI to interact directly with a user’s computer—clicking, typing, and managing files. The planned capabilities would position Gemini as a direct competitor to Anthropic’s Claude Cowork, which already offers similar agentic functionality for macOS users.
What the Update Includes
A teardown of the Gemini Mac app by 9to5Google uncovered evidence of upcoming agentic features. The AI is expected to:
- Control mouse and keyboard inputs – Execute clicks, scrolls, and keystrokes autonomously.
- Organize local files – Move, rename, or delete documents without manual intervention.
- Integrate with Google Workspace – Pull on-device content into Docs, Sheets, and other apps.
- Leverage macOS accessibility permissions – Use Screen Access and Accessibility APIs to interact with the desktop environment.
The Gemini Mac app, released on April 15, already supports screen sharing and local file uploads, but the new features would mark a shift toward full autonomy. Unlike the current version, which requires user approval for each action, the agentic update would allow Gemini to complete multi-step tasks independently.
How It Compares to Claude Cowork
Anthropic’s Claude Cowork introduced desktop control capabilities in late March, available to Claude Pro and Max subscribers. The feature allows Claude to:
- Open and edit files.
- Navigate web browsers.
- Manipulate UI elements (e.g., clicking buttons, filling forms).
- Operate even when the user is away from their computer.
Google’s approach appears to mirror this functionality, though the company has not yet confirmed whether Gemini’s agent will support offline operation. The Gemini 2.5 Computer Use model, released in October 2025, demonstrated strong browser and mobile UI control but was not optimized for desktop OS-level tasks. The macOS update would close that gap.
How to Prepare
Users interested in testing Gemini’s agentic features should:
- Install the Gemini Mac app – Available via the Google Workspace Updates blog.
- Grant accessibility permissions – Enable Screen Access and Accessibility in macOS System Settings (Security & Privacy > Privacy).
- Monitor Google I/O announcements – The feature may debut at the May 19–20 event, where Google is expected to unveil further agentic AI updates.
Tradeoffs and Considerations
- Privacy and security – Granting an AI agent control over a computer introduces risks, including unintended file modifications or exposure of sensitive data. Google has not detailed how it will mitigate these concerns.
- Subscription requirements – While the Gemini Mac app is free, agentic features may require a paid tier, similar to Claude’s model.
- Reliability – Early agentic AI tools can struggle with complex workflows or unexpected UI changes. Users should test the feature in non-critical environments first.
What to Watch Next
Google I/O 2026, scheduled for May 19–20, is the most likely venue for an official announcement. A keynote session titled Defining the agentic AI era suggests the company will double down on autonomous AI tools. If the macOS agent launches, expect Android and Windows versions to follow, along with deeper integrations into Google Workspace and ChromeOS.
For now, users can experiment with Gemini’s existing screen-sharing and file-upload features to gauge its potential. Those already using Claude Cowork may find Google’s offering a compelling alternative—once it arrives.