Grafy News
Try Grafy free

Google adds Mac‑wide computer‑use agent to Gemini

Google launched a Gemini desktop update that lets the AI directly operate macOS apps.

Illustration for the news story: Google adds Mac‑wide computer‑use agent to GeminiSource: TechCrunch — AI

Google announced on Thursday that its Gemini desktop application for macOS can now act on a user’s computer, navigating native apps, filling out forms and performing multi‑step workflows without the user having to click each button. The rollout, detailed by TechCrunch, marks Google’s entry into the growing market of “computer‑use” AI agents that can directly manipulate files, windows and web pages on a user’s behalf.

What Google announced for Gemini on macOS

According to TechCrunch, the new capability is built into the Gemini desktop client on macOS and is described as a “computer use” feature. After a user tells Gemini what they want to accomplish, the AI runs the steps in a live preview that appears inside the chat window, while the user’s cursor and screen remain free for other work. Users retain control: they can pause, stop or explicitly approve each action before it is executed.

How the Mac‑agent works in practice

The agent operates by issuing commands to native macOS applications and to web browsers running in the background. For example, a user could ask Gemini to "organise all PDFs from the last month into a folder called 'Reports'"; Gemini would locate the files, create the folder and move the documents. Similarly, the AI can sign into a website, complete a purchase, or book an appointment, all while showing a step‑by‑step preview in the chat. The preview lets the user verify the intended actions before they are committed, mirroring the safety controls seen in other desktop AI agents.

Typical workflows the agent can handle

Google suggests a range of use cases for the feature, focusing on repetitive or tedious tasks. These include:

  • Filling out repetitive online forms (e.g., expense reports, surveys)
  • Applying the same operation to a batch of files, such as renaming or converting formats
  • Sorting and organising files into folders based on user‑defined criteria
  • Generating slide decks from a collection of images, PDFs or data dashboards
  • Pulling information from web dashboards and compiling it into a report The live preview allows users to keep an eye on progress while continuing other work, which could be especially valuable for creators juggling multiple projects.

How Gemini’s agent compares with existing desktop AI agents

The announcement comes after competitors like OpenAI, Anthropic and Meta have already shipped similar agentic capabilities in their desktop apps. Google’s entry is later than those releases, but it integrates with the broader Gemini ecosystem, including the recently expanded “Tasks” feature that helps the model sign into sites, make purchases and schedule appointments. Early testers of Google’s revamped "CC" AI agent—introduced in September—have been using it to coordinate email, calendar and chat across families. The new computer‑use ability extends that coordination onto the local machine, giving Gemini a more hands‑on role in the user’s workflow.

Trying similar agentic workflows with Grafy

While Grafy does not yet host Gemini, creators and developers can explore comparable agentic concepts using existing Grafy tools. The Grafy Operator app functions as a browsing agent that can perform web‑based actions on a user's behalf, providing a sandbox for testing automated workflows such as form filling or data extraction. For multi‑model experimentation, Grafy Chat lets you run several AI models side by side, enabling you to compare how different agents interpret the same instruction. If your workflow involves documents, the Grafy Reader can ingest PDFs and let an AI answer questions about their contents, mimicking the information‑gathering step of a Gemini task. By chaining these apps—e.g., using Operator to fetch data, Reader to analyse it, and Chat to orchestrate the steps—you can prototype end‑to‑end automation without needing Gemini itself.

Limitations and privacy considerations

The feature is currently limited to English users in the United States and requires a paid Google AI subscription (Google AI Plus, Pro or Ultra). Google has not announced when the capability will roll out to other regions or platforms. Because the agent can manipulate files and interact with web services, users must trust that the AI respects privacy settings and does not unintentionally expose sensitive data. Google’s pause/stop/approve controls are intended to mitigate accidental actions, but developers should still audit any automated workflow for security implications before deploying it at scale.

Source: TechCrunch

Frequently asked questions

Does Gemini’s new Mac agent automatically click buttons without user confirmation?

No. The agent presents a live preview of each step in the chat window, and users must explicitly approve, pause, or stop actions before they are carried out on the computer.

Is the computer‑use feature available on Windows or Linux?

According to the announcement, the capability is limited to the Gemini desktop app on macOS and is only released in English for users in the United States. No Windows or Linux support has been mentioned.

Can I try a similar AI‑driven automation without a Google subscription?

Grafy offers comparable automation through its Operator and Chat apps, which are available to all Grafy users. While they don’t replicate Gemini’s exact integration with macOS, they let you build and test multi‑step AI workflows on web content and documents at no extra cost.


Source: TechCrunch — AI — Google gives Gemini the ability to use your Mac for you