Google Is Turning Its AI Assistant Into a Robot That Uses Your Computer

Google has reorganized the team working on Project Mariner, an AI system that can move a computer mouse, click buttons, and fill out web forms by itself. The shift marks a turning point: Google is folding this experimental technology into its main products rather than keeping it as a lab project.

Over the past few years, Google built tools to let an AI understand what's on your screen and interact with websites the way you would. Now the company is taking those capabilities and building them into its Gemini 2 chatbot and its Chrome web browser. The goal is to let you tell an AI to do routine computer tasks—booking a flight, managing your email, researching something—without you having to click every button yourself.

What Is an AI Agent, and Why Does Google Care?

Think of an AI agent as a digital assistant that can operate a computer much like you can. You give it a goal in plain English: "Book me a flight to Portland next Tuesday." The AI then figures out what steps to take, visits the websites it needs, and completes the task on its own.

Google's Gemini 2 is the company's new model trained specifically to work this way. Alongside it, Google released two AI agents focused on specialized tasks: one for writing and fixing computer code, and another for data analysis. Both use Gemini 2's ability to understand complex instructions and break them into steps it can execute.

Project Mariner, which started as a small experiment, is the technical foundation for all of this. Mariner was built as an extension for Chrome (Google's web browser) and learned to recognize buttons, text fields, and other objects on web pages, then interact with them on behalf of the user. The technology is now the backbone of Google's bigger agent strategy.

Building Agents Into the Browser

Google has also added new AI features directly into Chrome itself, including a button that opens the Gemini chatbot right from your browser. These tools started rolling out to paid Google users in May but are now available to everyone with a Chrome browser on a desktop computer in the United States who reads content in English.

The logic here is straightforward: your web browser is where you spend most of your work time online. By putting AI agents into Chrome, Google positions itself to help you automate tasks at the moment you're already working. This follows a pattern Google has used before—putting search into Gmail, putting Google services everywhere on Android—to embed itself into tools people already use daily.

Technical Progress and Real-World Use

Google's decision to move Mariner's technology out of the lab and into actual products suggests the company has solved a real problem: making AI reliable enough to handle the unpredictable nature of the web. Browser automation has always been tricky because websites change constantly, and AI systems need to keep up.

The shift also tells us something about Google's confidence. The company is no longer treating AI agents as something to research indefinitely. It's betting that people actually want to use them for everyday tasks—paying bills, planning trips, managing paperwork—and that it can turn this into a business.

For offices and businesses, this opens new possibilities. Companies could theoretically use natural language to automate routine computer tasks that normally require custom software or manual work. This could save money and time, though it would require companies to trust an AI system to handle sensitive operations.

The Broader Picture

Other major technology companies are pursuing similar capabilities. Anthropic (which makes a chatbot called Claude) recently added computer use features, and Microsoft is integrating AI agents into its Copilot system. The shift is happening across the industry.

The question facing Google and its competitors is whether people will actually trust an AI system with computer access. An AI agent that has gone wrong could delete files by mistake, make unwanted purchases, or access information it shouldn't. Google's approach of building these features into existing products—rather than releasing a standalone agent—suggests the company is thinking carefully about safety and user control.

What happens next depends partly on whether people find these tools useful enough to adopt, and partly on whether Google can keep humans in control while letting the AI do its work. The technical pieces are falling into place. Whether this actually becomes part of how people work remains to be seen.