Google recently introduced a new feature called "Auto Browse"（自动浏览），which transforms the Chrome browser into a more autonomous AI platform and competes head-to-head with products like ChatGPT. This feature is based on Google's Gemini system, aiming to shift routine operations such as filling out forms, collecting data, and planning trips from users' hands to digital proxies.
Google says that Auto Browse is one of the most important steps towards "agentic AI"（代理式AI），a system with contextual understanding capabilities that can automate user interface interactions. Currently, this feature is in preview mode and only available to paid subscribers, including AI Pro and AI Ultra plans, which run on the latest Gemini 3 model architecture influenced by Google's early experimental proxy framework Project Mariner.
When a user starts an Auto Browse task, Chrome opens a special tab page with dynamic AI icons. This proxy can jump between websites, follow links, and interact with form fields through simulated keyboard and mouse input without requiring constant attention from the user. The system will notify the user when tasks are completed; if sensitive operations like accessing passwords or initiating payments are involved, users will receive pop-up requests for permission.
Auto Browse adds a new abstraction layer to tools like Atlas from OpenAI, allowing users to describe their intent in natural language and Gemini to break it down into a series of operations and execute them step-by-step. This enables multiple backend tasks to run concurrently, reducing the cognitive load on users when performing repetitive web page operations.
The Chrome interface has also been adjusted around Auto Browse. Gone are static Gemini buttons now defaulting to a side panel, which lets AI access current page content continuously. In this side panel, Gemini can coordinate with other Google services like Gmail, calendar, maps, YouTube, shopping, and flights in real-time, achieving cross-service synchronization.
Google has also introduced Nano Banana image editing functionality within the browser, allowing users to edit images without downloading files directly. For visual tasks, users can switch between standard models for speed and higher-quality Gemini Pro for quality. However, Auto Browse relies heavily on cloud: all proxy operations on websites are transmitted in real-time to Google servers, where Gemini models handle them.
According to Google's explanations, web content may be temporarily recorded and stored under user accounts in Gemini Apps Activity data, depending on their privacy settings. It is unclear whether the accessed content in Auto Browse sessions will be used to train future AI systems.
In terms of entry barriers, AI Pro subscribers can start 20 tasks per day, while AI Ultra users have 200 session quotas per day. Although currently preview features do not require an additional purchase, Google has not announced when they will open access for free users. To prevent misuse risks, Auto Browse is set with protection rules that prohibit it from executing transactional operations without user confirmation, such as directly submitting payment information.
In practice, auto browsing can recognize purchased goods and fill in order forms but stops at the final checkout step, waiting for user confirmation. Whether Auto Browse will transition from experimental preview to mainstream application depends on users' trust in letting a proxy represent their online actions and behavior in the web world.
Google says that Auto Browse is one of the most important steps towards "agentic AI"（代理式AI），a system with contextual understanding capabilities that can automate user interface interactions. Currently, this feature is in preview mode and only available to paid subscribers, including AI Pro and AI Ultra plans, which run on the latest Gemini 3 model architecture influenced by Google's early experimental proxy framework Project Mariner.
When a user starts an Auto Browse task, Chrome opens a special tab page with dynamic AI icons. This proxy can jump between websites, follow links, and interact with form fields through simulated keyboard and mouse input without requiring constant attention from the user. The system will notify the user when tasks are completed; if sensitive operations like accessing passwords or initiating payments are involved, users will receive pop-up requests for permission.
Auto Browse adds a new abstraction layer to tools like Atlas from OpenAI, allowing users to describe their intent in natural language and Gemini to break it down into a series of operations and execute them step-by-step. This enables multiple backend tasks to run concurrently, reducing the cognitive load on users when performing repetitive web page operations.
The Chrome interface has also been adjusted around Auto Browse. Gone are static Gemini buttons now defaulting to a side panel, which lets AI access current page content continuously. In this side panel, Gemini can coordinate with other Google services like Gmail, calendar, maps, YouTube, shopping, and flights in real-time, achieving cross-service synchronization.
Google has also introduced Nano Banana image editing functionality within the browser, allowing users to edit images without downloading files directly. For visual tasks, users can switch between standard models for speed and higher-quality Gemini Pro for quality. However, Auto Browse relies heavily on cloud: all proxy operations on websites are transmitted in real-time to Google servers, where Gemini models handle them.
According to Google's explanations, web content may be temporarily recorded and stored under user accounts in Gemini Apps Activity data, depending on their privacy settings. It is unclear whether the accessed content in Auto Browse sessions will be used to train future AI systems.
In terms of entry barriers, AI Pro subscribers can start 20 tasks per day, while AI Ultra users have 200 session quotas per day. Although currently preview features do not require an additional purchase, Google has not announced when they will open access for free users. To prevent misuse risks, Auto Browse is set with protection rules that prohibit it from executing transactional operations without user confirmation, such as directly submitting payment information.
In practice, auto browsing can recognize purchased goods and fill in order forms but stops at the final checkout step, waiting for user confirmation. Whether Auto Browse will transition from experimental preview to mainstream application depends on users' trust in letting a proxy represent their online actions and behavior in the web world.
2025年1月10日，Google推出新功能"自动浏览"，Chrome浏览器成为更加自主的AI平台。 
根据 Gemini系统，自动浏览可以让用户不需要 manuallyfill out表格、收集数据和计划旅行，直接交给digital proxies处理。 
根据Google的说法，这是他们走向"代理式AI"（agentic AI）的重要一步。 
但实际上，现在该功能仅在付费订阅者中试验，包括AI Pro和AI Ultra计划，仅适用于最新的Gemini 3模型架构。
此外，自动浏览功能已在试验期内获得90%的用户满意度。 
根据测试结果，20%的实验者表示他们在使用自动浏览功能时感受到了不适感。 
️ 我們看到 Gemini System 這個架構被 Google intensively 運用，可以讓 Chrome 轉型成更加的 AI 平台。 
但是在我看來，使用者必須要小心，不要讓 Auto Browse 的 proxy 這個功能過度依賴於 cloud，否則會影響到 my online Privacy 
這個功能也給我們一些新的想法，例如為了方便的線上購物，我們可以使用 Auto Browse 進行購物流程，填寫購物信息不用手動打字，可以節省時間。 
但是，我們應該要小心，不要讓 Auto Browse 的 proxy 這個功能開始對我來說過度依賴，否則會影響到 my line life 
我們看到 Google 已經引入了 Nano Banana image editing 功能，這是 Pretty cool 的功能，可以方便地在線上編輯圖片。 
，但是感觉一下子就感到像是在放心掉了？ 
个人觉得应该加上一个开启/关闭的选项，让用户能够更好地控制这个功能。另外，如果没有一些保护措施，我认为这种功能会让更多的人去尝试，所以需要在用户同意后完成一些敏感的操作，才不会让人担心。 
！Auto Browse 的说法让我感到很有意思，感觉和 ChatGPT 一样的！
2025, browser just not just for browsing anymore, need to automate task, like filling out forms, collecting data... 
Google saying that web content may be recorded in Gemini Apps Activity data, but what if they use it for training AI? 
️ all proxy operations are transmitted to Google servers, where Gemini models handle them. 
How they going to access this feature? 
。
我们还能让 AI 代表我们在网上做事吗？ 
这个Auto Browse功能我觉得还是有点让人担心 
or we can just trust Google to protect our data?
自动化能力和 AI 的融合，让我们的浏览体验变得更加智能化 
那么Google会不会考虑降低限制？ 
会不会这些存储的用户行为数据会被用于其他目的？ 
but 我也担心一下，如果我们真的放手给机器去做这些事情，我们到底要保证哪些安全性？ 
毕竟这是在网上，我们都知道有很多风险 
和bug 
，所以我希望google能在推出后继续加强一些安全性的控制措施 
我觉得，这种功能还需要更好的隐私保护机制，特别是在处理敏感信息的时候。 
不過，它也可能让我们失去对自己的控制权呢？ 