When AI can complete most complex tasks with a box and a browser interface, it also means that the only task left on the desktop is to support people to open AI browsers.
When people enter the Internet era, they have embarked on a relatively linear development path of web pages, browsers, portals, searches, and social networks around the needs of information production, sharing, and consumption. Standing on this path, in the process of embracing the AI era, different enterprises and developers use this as a reference, and at the same time look for the possibility of AI at different nodes.
Among them, the browser is the first national application in the Internet era. Bill Gates wrote in his famous memorandum “Internet Wave” released thirty years ago: We believe that in the next 20 years, computers will have a fundamental impact on work, study and gaming. Great software products are essential to achieving these advancements. Microsoft began to compete for the browser as an entrance software.
In the AI wave, AI browsers have become one of the most competitive nodes in AI applications. There are many competitors in AI browsers, from star entrepreneurs to Internet veterans, from domestic giants to overseas rookies. Fellou, Dia, and Opera are all looking for the best implementation path for native AI browsers, Chrome and QQ browsers have also embedded AI assistant capabilities, and OpenAI and Perplexity are also releasing their own browser products.
To achieve these three challenges, product managers will only continue to appreciate
Good product managers are very scarce, and product managers who understand users, business, and data are still in demand when they go out of the Internet. On the contrary, if you only do simple communication, inefficient execution, and shallow thinking, I am afraid that you will not be able to go through the torrent of the next 3-5 years.
View details >
Entrance is the most direct temptation that browsers give people. Every change of mainstream browser is a change of hands. Microsoft and Netscape launched the first platform battle in the Internet era around browsers, and Microsoft’s victory allowed it to further strengthen its dominance on the desktop through the combination of Windows + IE. When Google launched Chrome, it overturned the IE browser lying on the credit book to the ground, and also took advantage of the trend to find the most suitable partner for search.
When people realize that AI may bring new disruptions, browsers are still one of the core components of the AI experience. However, the AI browser has changed from a search container to a body of AI capabilities, from a window for obtaining information to a partner for proxy execution. Behind this change, AI is integrating search, decision-making, and execution, allowing people to complete more complex operations through natural language.
This also makes AI browsers an overlapping point, where AI applications in different directions such as AI search, general agents, and copilots will intersect. As AI assistants, Doubao, AI search product Quark, and knowledge base product ima all choose to carry their product capabilities in the form of browsers. At the same time, the virtual computer configured by Buttonspace and Manus for AI is more like a collection of web tools based on browser capabilities.
Under this trend, AI browsers will become more like an operating system for the AI era. This is what Google had expected Chrome to achieve – bypassing the traditional desktop system and using the browser as the core to build a new software and hardware ecosystem using cloud capabilities.
When AI can complete most complex tasks with a box and a browser interface, it also means that the only task left on the desktop is to support people to open AI browsers.
In AI browsers, we see a simpler future interaction paradigm between people and the world. This interactive paradigm will not only change the software, but also make the hardware a local computing power provider and cloud computing power connector for AI, creating a more flexible and lightweight software and hardware ecosystem. The former can make full use of computer hardware configurations to support AI on browsers to run complex tasks, while the latter can perform simple tasks on lightweight devices such as smart glasses.
01 AI-powered browser makeover
Marc Andreessen, founder of Netscape, once said that browsers are the software that has had the greatest impact on people’s lives in history. This judgment has not changed today. Even today, when the mobile Internet is so popular and people are accustomed to searching and browsing information on WeChat and Douyin, browsers still have hundreds of millions of users around the world, undertake a large part of complex information acquisition work, and are the main way to open shared documents and multi-dimensional tables.
When AI becomes the driving core of a new wave of technology, browsers have also become a key carrier of AI capabilities. Many AI browser products that have been released are relying on smarter search experiences and more complex task intelligence capabilities to gain attention and pursuit, creating a scene of “hard to find” again and again. Up to now, if you want to register and use AI native browsers like Fellou and Dia, you still need to ask for an invitation code first.
Under the premise of AI-driven, although different AI browsers follow the design of AI dialog + browser interface, they will have different emphasis on specific use. Their different positions and starting points determine the development logic and user experience of the product.
We can roughly divide AI browsers into three categories: AI-native browsers, AI-like browsers, and AI-based traditional browsers.
Among AI native browser products, Dia makes me feel more like I’m using a browser, and the operation is easier and smoother. After filling in the content in its search box, users can choose to search to open a web page or start a chat conversation. In the already opened page, users can also start AI conversations at any time to summarize and ask questions on one or more pages, and the workflow is to meet the needs of information acquisition before further transitioning to writing articles or writing code.
In comparison, Fellou feels more like a browser-like Manus with more complex functions. Because of its emphasis on agent-based deep action and proxy browser, when I use it, I immediately have a mentality of assigning tasks to my assistants. After the task is issued, wait for the assistant to complete the task. This feeling of use is completely consistent with Manus and button space, and there will be no strong browsing experience.
In addition to AI native browser products, the desktop versions of Doubao, Quark, and ima also exist in the form of browsers, but they will not highlight the positioning of the browser in the product positioning. For these products that focus on AI assistants, AI search, and AI knowledge bases, the browser is more like a product function that comes with users, allowing users to open task pages in it, and their product upgrades will still focus on the optimization of AI assistants, search, and knowledge bases.
In addition, traditional browsers are also seizing the opportunities brought by AI to achieve the goal of sprouting old trees. QQ Browser has launched QBot, which supports multi-modal search, automatic recognition of page browsing intent, and auxiliary call of office capabilities. Chrome also integrated Gemini AI assistant in the update, and Microsoft launched the AI assistant Copilot Vision for the Edge browser. This change actually adds an AI plugin to the browser, giving it more AI capabilities beyond the original experience.
Compared with AI-native browsers and AI-like browser products, traditional browsers will have more baggage when carrying out AI, making it difficult to quickly switch an old product to a new track. Therefore, they often choose to transform their thinking gradually. Liu Hanbo, head of QBot products, said that QQ Browser will achieve incremental improvement through AI on the basis of retaining the original high-quality business content. AI has indeed brought a certain increase. In the grayscale testing stage before the official release of the QQ browser AI function, the activity of the covered users has increased, and the entire market is going up.
02 Browser Competition Logic: Entrance, Operator and Smart Partner
Although the implementation ideas of the three types of products are different, the competition logic behind them is the same.
First of all, the competition around AI browsers is to compete for a new round of entrance control, which is already a consensus. Josh Miller, CEO of Dia’s developer, The Browser Company, said in an open letter that cloud revenue has skyrocketed, emerging startups are using browsers as a platform, the cryptocurrency ecosystem relies on browser plugins, and WebAssembly opens a new interactive experience……
Everything is converging in the browser.
This aggregation trend has become more pronounced under the stimulation of AI. People are increasingly relying on the web to achieve AI search, AI image generation, AI video generation and other needs. At the same time, we can also see significant growth in AI social products on the web in the first quarter of 2025; E-commerce platforms such as Taobao have also begun to refocus on PC experience; People interact with agent products such as Buckle Space and Manus more on the web page.
At the same time, AI is changing the way people match their needs in the browser. While traditional pages are still relied on to carry these capabilities, people may not need to see traditional pages in the future. More requirements will be raised in the browser’s address bar, and then the AI will assign and execute them, and finally feed back the results in the chat interface extended by the address bar. Under such a development trend, the browser’s address bar will also evolve into a super box like Quark’s search box, becoming the core entrance of AI.
Second, the competition is no longer just the entrance to information, but also the starting point of the automated task flow, where the browser will play the role of an operator.
Perplexity CEO Arvind Srinivas believes that Perplexity’s products are a series of workflows, such as Perplexity’s Agent can help users manage their portfolios, and after reading reports and news, it will tell users how to adjust their risk exposure. This agent needs to understand the entire search and information browsing process, not just a single term search.
The browser is the right carrier to carry this process. Arvind Srinivas said that if you really want AI to move from providing answers to taking action and doing things for you — an answer is essentially a combination of four or five searches, and an action is a full browsing session, done with a single prompt — you need to have a browser and seamlessly integrate computing power on both the client and server sides.
In the introduction of Fellou founder Xie Yang, the top four use cases of Fellou products are information query, data crawling, social media posting and automatic form filling, and 60% of the use cases are data query and crawling. Someone sent a job application message to his favorite company on LinkedIn through Fellou, and finally received interview invitations from 10 companies and successfully joined one of them. This shows that users do expect AI browsers to have stronger action.
It should be noted that AI browsers can obtain more contextual information about the user through cookies, such as what the user has browsed on the website, what he has done, and what account he has logged into. At the same time, AI browsers allow users to access other third-party services through hidden tabs while logged in, scrape page data, and reason and perform actions on behalf of users. This solves the problem that AI applications have no underlying control and cannot directly call other applications and obtain relevant information.
Finally, the browser battle may ultimately be a battle for smart partners.
Browsers will be part of AI assistants in the future. Xie Yang believes that what Fellou does is to give more people freedom; Let more people have time; Let more people be themselves. Josh Miller told The Verge that Dia’s core function is a chat tool that helps users find information and get things done more clearly and easily.
These positions all point to an intelligent partner with more memory and execution.
03 Continuation of the Chrome OS story
From my own experience, the current AI browser is not a competent intelligent partner, but more like a child who has not yet learned to use tools and professional skills.
We can ask AI browsers to write emails, resumes, and web pages on our behalf, but in many cases they can only reach the passing level, and even some complex tasks will collapse during execution. For example, when using Fellou to create a news newsletter of the day, it struggles to get enough valid information from a news site.
AI browsers need to have a capability and tool ecosystem. To some extent, the main way for AI browsers to obtain information is no longer to browse the web, but to communicate directly with different applications and service providers at the data or protocol level. Neon, an AI browser developed by Opera, does not use page browsing to obtain information, but analyzes web content through DOM tree and typesetting data to obtain effective information more quickly.
This ecology may be divided into two layers. The first layer is based on the capabilities of AI browsers, which can create a new AI plug-in system. Dia is trying to build such a skill ecosystem, allowing users to use Dia’s “skills” function to write code, create different tasks, abilities, and distribute them like the App Store, becoming a “routing system” that matches needs and capabilities.
The second layer is to open up with third-party platforms. The Information said in a November 2024 report that OpenAI is also building an AI browser product that will be deeply integrated with ChatGPT and will be connected to the capabilities of service providers such as travel, food, real estate, and retail. When it comes to the shopping agent that Amazon is trying to build, we think that the access of this third-party capability will most likely be achieved through the interaction between agents.
AI browsers with an application ecosystem will be more like an operating system. Josh Miller made it clear that Dia will ultimately be a browser-based system, and that this system will be a powerful command center in the future full of agents. Arvind Srinivas also emphasized that browsers are essentially containerized operating systems and are the best way to build agents.
It’s a lot like an AI upgrade of Chrome OS Stories. Google’s Chrome OS, launched in 2011, carried its ambition to develop an operating system based on web applications and cloud services. Since then, Google has also launched Chromebooks with Chrome OS. Chromebooks are not equipped with Windows systems through low-configuration configurations like netbooks to achieve light office work locally, but use the combination of cloud resources and Google services to meet the daily use needs of users.
The earliest Chromebooks were complained about as “browsers with shells”. But with the continuous iteration of Chrome OS, Chromebooks have continued to conquer the education market in North America and Europe, becoming the main choice for notebook consumers under $350. Taking this as a reference, if AI browsers really become people’s main computing space, AI browsers may also have hardware products that match them.
This hardware can be a PC with local computing power or smart glasses that rely only on cloud computing power. From the perspective of seamless connectivity, AI Browser is a lightweight AI operating system similar to Chrome OS, creating a new ecosystem of software and hardware around AI, allowing people to live on the AI Internet.
