Google is quietly transforming the Chrome browser into a more autonomous AI platform through a new feature called "Auto Browse" and competing head-on with products such as ChatGPT. This function is based on Google's Gemini system, and its goal is to transfer routine operations such as filling out forms on web pages, collecting data, and planning trips from users to digital agents.
Google says Auto Browse is one of its most important steps toward "agent AI," a system that can automate user interfaces through reasoning with contextual understanding. Currently, this feature is in preview, only open to paid users, including AI Pro and AI Ultra subscription plans, and runs on the latest generation Gemini 3 model, whose architecture is influenced by Google's early experimental agent framework Project Mariner.
When a user launches an Auto Browse task, Chrome opens a dedicated tab identified by a dynamic AI icon. The agent can jump between websites, follow links, and interact with form fields by simulating keyboard and mouse input. Users do not need to keep an eye on the progress throughout the process, as the system will issue a notification when the task is completed; if sensitive operations such as access passwords and payment initiation are involved, a prompt requesting user authorization will pop up midway.
Similar to OpenAI’s Atlas tool, Auto Browse adds a new layer of abstraction between user intent and web interfaces. Users no longer directly control the browser, but describe goals in natural language, and Gemini breaks down these goals into a series of operations and executes them step by step. This approach allows multiple background tasks to run in parallel, aiming to reduce the user's cognitive load on repetitive web page operations.
Around Auto Browse, Chrome's interface has also been adjusted. The static Gemini button in the past now resides in the form of a sidebar (Sidepanel) by default instead of a pop-up window, allowing AI to continuously access the current page content. In this sidebar, Gemini can coordinate Google services such as Gmail, calendar, maps, YouTube, shopping and air tickets directly within Chrome to achieve cross-service linkage.
Google also introduced the Nano Banana image editing function in the browser, allowing users to edit images locally without downloading files. When working on visual tasks, users can switch between the faster standard model and the higher-quality Gemini Pro to balance efficiency and quality. However, the operation of Auto Browse is highly dependent on the cloud: all interactive operations performed by the agent on the web page will be transmitted to the Google server in real time and processed by the Gemini model.
According to Google, web content may be temporarily recorded under the user's account and saved in Gemini Apps Activity data, depending on the user's privacy settings. Google has not yet clarified whether the content accessed in Auto Browse sessions will be used to train future AI systems.
In terms of usage thresholds, AI Pro subscribers can initiate up to 20 browsing tasks per day, and AI Ultra users can receive a daily session quota of 200 times. Although the current preview feature requires no additional purchase, Google has not announced when it will open access to free users. In order to reduce the risk of misuse, Google said it has set protection rules for Auto Browse to prohibit it from performing transactional operations without explicit confirmation from the user, such as directly submitting payment information. In actual use, automatic browsing can identify the goods to be purchased and fill out the order form on its behalf, but it will stop before the final checkout step and wait for the user's personal confirmation.
Whether Auto Browse can move from experimental preview to mainstream use still depends on whether users trust such an agent enough to let it browse and act on their behalf in the online world.