Gemini for Mac Learns to Click Through Apps and Fill Out Forms on Its Own
Google has taught the Gemini app for Mac to click through programs, type, and fill out forms by itself. You describe the task in plain words, and the assistant carries it out in a background window, without taking over your screen or cursor. For example, it can move weights and addresses from a PDF into a delivery cost calculator. Google warns of errors that can go as far as deleting files, so it is best to start with a test folder.
AI-processed from TNW; edited by Hamidun News
Agents that can use a computer the way a person does have long stopped being conference demos. Now they are arriving in ordinary consumer apps. Google has announced that computer use mode is rolling out in the Gemini app for Mac: the assistant takes a task and works its way through local files, desktop programs, and websites, clicking, typing, and filling in forms.
How It Works
Users describe the task in ordinary words. Gemini then carries it out in a separate background window, and the chat shows a live preview of what is happening. As a result, your work screen and cursor stay free: you can keep writing a document or answering email while the agent handles its job.
People stay in control. You choose which apps Gemini may use, you can pause the task, and the agent must get your approval for key actions. In addition, it sees only the folders you have connected to it, not the entire disk.
Google gives several everyday examples. The first: pull package weights and addresses from PDF files and enter them into a delivery cost calculator. The second: turn data from a dashboard into a presentation in Google Slides. The third: edit photos in batches. The fourth: prepare resale listings for items based on photos in a folder. What all these tasks have in common is that each consists of dozens of small, routine steps across different programs, which is exactly the kind of work people usually don't want to do by hand.
Who Gets Access and What the Limits Are
The feature is available only to paid subscribers on the AI Plus, Pro, and Ultra plans. It requires a Mac with an Apple Silicon chip, so older Intel-based computers are left out. As with several other Google features, the mode is not available in Europe or the UK, so users there will have to wait or rely on other tools.
The company openly warns that Gemini can make mistakes. Among the possible consequences, it explicitly names the permanent deletion of files and sending them to other people. To reduce the risk, the agent is designed to ask for permission first. Backups of files it has changed are kept only briefly: they are deleted after 24 hours or when you start a new task. This means you can undo a bad edit only within a short window, and it is worth keeping that in mind.
What It Means in Practice
For product managers and founders, the main signal is elsewhere: the race for desktop agents is reaching the mass market. Google is building such an assistant into an app that ordinary people use, and it is betting on background work instead of taking over the screen. According to the source, competitors are moving in the same direction too: ChatGPT on Mac can already work with local apps to feed its memory, and agents built on the OpenClaw framework have noticeably fueled interest in the topic.
Developers should look closely at the security approach. An allowlist of apps, connected folders, confirmation of critical steps, and a short backup retention period form a set of practices that will be useful in any custom agent with access to a computer. If you are building something similar, it is better to put this baseline in place from the start.
For an ordinary user, a sensible approach is not to give the agent working directories that hold the only copies of your documents. Set up a test folder with copies of your files, ask Gemini to handle a routine task such as moving data from a PDF into a form, and see how often it makes mistakes and how clearly it asks for confirmation. Only after that does it make sense to give it more serious tasks.
For now, this is an early launch with caveats from Google itself. But the direction is clear: the assistant is no longer just a window for questions and is starting to do the work in the same programs where you do it yourself.
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026MIRA Neural Network Renders Rocket League at 20 FPS, but Ping Hits 460 ms
- Nov 21, 2026Google built a closed copy of the internet with fake banks to train Chrome's AI agents
- Nov 21, 2026MLPerf Training Adds Nvidia Rubin and TPU Ironwood in First Appearance: 153 Systems
Key Questions Answered
How It Works?
Users describe the task in ordinary words. Gemini then carries it out in a separate background window, and the chat shows a live preview of what is happening. As a result, your work screen and cursor stay free: you can keep writing a document or answering email while the agent handles its job. People stay in control. You choose which apps Gemini may use, you can pause the task, and the agent must get your approval for key actions. In addition, it sees only the folders you have connected to it, not the entire disk.
Who Gets Access and What the Limits Are?
The feature is available only to paid subscribers on the AI Plus, Pro, and Ultra plans. It requires a Mac with an Apple Silicon chip, so older Intel-based computers are left out. As with several other Google features, the mode is not available in Europe or the UK, so users there will have to wait or rely on other tools. The company openly warns that Gemini can make mistakes. Among the possible consequences, it explicitly names the permanent deletion of files and sending them to other people.
What It Means in Practice?
For product managers and founders, the main signal is elsewhere: the race for desktop agents is reaching the mass market. Google is building such an assistant into an app that ordinary people use, and it is betting on background work instead of taking over the screen. According to the source, competitors are moving in the same direction too: ChatGPT on Mac can already work with local apps to feed its memory, and agents built on the OpenClaw framework have noticeably fueled interest in the topic.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.