Without much fanfare, Google is giving Chrome a new role. Auto Browse, powered by Gemini AI, nudges the browser toward acting on users’ behalf, thinking through context, and taking control of on-screen tasks. It’s a subtle shift, but one that puts Chrome in a collision competition with AI assistants like ChatGPT.
For now, Auto Browse is limited to a preview for Google’s AI Pro and AI Ultra subscribers, running on the latest Gemini 3 models. The feature borrows from Project Mariner, an earlier agent experiment, and points toward a future where mundane tasks like filling forms, collecting information, or organizing travel plans are handled by a digital assistant instead.
Once a task starts, Chrome opens separate tabs identified by an animated AI icon. The agent moves between sites, follows links, and fills in form fields by simulating keyboard and mouse actions. Users can step away, with notifications appearing only when a task is finished or when approval is needed for sensitive steps such as password access or payments.
Auto Browse works a lot like OpenAI’s Atlas, putting some distance between what you want to do and the browser’s buttons and menus. You tell it your goal, and Gemini figures out the steps and handles them for you. That setup lets several tasks run in the background at the same time, cutting down on the mental effort of doing the same browsing chores over and over.
To make this work, Chrome’s interface has been reworked. The Gemini button, which used to just sit there, now opens into a side panel that stays on screen and keeps track of the page you’re on. From that space, Gemini can tap into services like Gmail, Calendar, Maps, YouTube, Shopping, and Flights, all without bouncing between tabs.
The update brings image editing directly into the browser through Nano Banana, removing the extra step of downloading files. When working with visuals, users can opt for speed with the standard model or switch to Gemini Pro for more detailed results.
Auto Browse operates entirely in the cloud, with agent-driven page interactions streamed to Google’s servers for Gemini processing. Documentation indicates that page content may be temporarily logged to a user’s account and stored in Gemini Apps Activity, subject to privacy preferences. Google has not confirmed whether Auto Browse data is used to train future AI models.
Auto Browse includes safeguards that restrict transactional actions without direct user authorization. The agent can surface purchase options and populate order forms, but it is designed to halt before executing a final checkout.
The preview comes with defined usage limits. AI Pro subscribers are capped at 20 Auto Browse sessions per day, while AI Ultra users can initiate up to 200. No additional purchase is required, though Google has not disclosed plans to extend access to free-tier accounts.
Maybe you would like other interesting articles?