OpenAI's new browser saves me this Double 11.

Every year during the Double 11 shopping festival, it's like a boomerang hitting me, a person who graduated from high school years ago - I simply can't figure out the discounts. It's just too confusing!

But this year is different.

I asked ChatGPT Atlas (the AI browser just launched by OpenAI last week), "Should I buy the iPhone 17 or the iPhone 17 Pro? How can I get the best deal?"

Then, Atlas automatically retrieved the product pages I had browsed. It analyzed historical prices, ongoing promotions, and user reviews to summarize the pros and cons of each platform and the available discount schemes for me...

Today's commercial battles are best left to AI | Image source: ChatGPT Atlas

All these were completed within a single interface (then what was all the suffering I went through before???). It was so seamless that I almost forgot what a "specific" action "browsing the web" is - opening a search engine, typing in keywords, scrolling through the search results page by page, only to find that the content wasn't what I was looking for when I clicked on it.

I often have this sense of confusion, like "where am I being led?" | Screenshot by the author

If you want to compare the prices of the new iPhone during work breaks but can't use your phone freely, you end up opening a dozen tabs and constantly switching between several e - commerce platforms to compare prices, read reviews, and calculate discounts...

The information is messy, the web pages are fragmented, and the switching is chaotic - but we've gotten used to it.

"Little Smart Tricks"

Browsers have been getting smarter.

Since its launch in 2008, Chrome has brought in the golden age of modern browsers with faster loading speeds, a cleaner interface, and an open plugin ecosystem.

Chrome's open ecosystem has greatly enhanced its appeal | Screenshot by the author

In the following decade, browsers reached a peak of functional innovation, with features like tab grouping, reading mode, cross - device synchronization, and plugin markets...

Browsers have become more powerful, but one thing has remained unchanged: they only open the door but never care if you've found what you're looking for - because browsers don't understand what you're doing.

It wasn't until the emergence of AI that AI plugins almost overnight occupied the browser sidebar. You can instantly summon "ChatGPT - like" assistants on the right side of the web page to summarize the page or translate YouTube videos in real - time.

"AI plugins" have become a highly competitive "race track." The reason is simple: no one wants to miss the opportunity to seize the "user entry point."

AI plugins in the Chrome Web Store | Screenshot by the author

The experience improvement brought by AI plugins mainly lies in reducing the friction of jumping between multiple web pages. However, they are like "add - ons" and haven't integrated the complete user usage scenarios. For example, they can summarize a web page but don't know which question you started searching from.

In short, this kind of experience improvement is patchy. The "core" remains the same, and our way of surfing the web hasn't changed.

Microsoft was the fastest among the tech giants. Microsoft integrated Copilot into Edge, enabling it to perform functions that previously required various plugins. Edge now has a powerful "co - pilot," but the one driving the "car" remains the same - Edge hasn't become a completely new species.

So, what does a completely new species look like?

The Landscape of AI Browsers

AI browsers are subverting the primary - secondary logic. Instead of embedding AI into browsers, they are built around AI.

The starting point of this revolution begins with Arc.

In 2022, The Browser Company launched the Arc browser with the slogan of "redefining the web." It completely abandoned Chrome's old template, stuffing tabs, bookmarks, notes, and the download manager into the sidebar, allowing users to take notes while browsing.

Although it didn't have much of an AI flavor at first, Arc made users realize for the first time that a browser isn't just for viewing web pages but can be extended to the desktop system.

Arc made a grand entrance | Image source: Arc

In 2023, Arc started experimenting with AI, adding features like web page summarization and cross - tab search. Although it still relied on external models like GPT - 3, it showed the possibility of a browser "thinking for itself."

By the end of 2024, Arc announced that it would stop updates, and the team shifted to launching a brand - new AI - native browser: Dia. This is also considered the starting point of the "Year of AI Browsers."

In 2025, the AI browser market officially exploded: Dia, Comet, Zen, Genspark...

Let's start with Dia, which inherits from Arc. It directly turns the browser into the "command center of the Internet."

Dia radically removed the address bar and tabs, not intending for you to "browse on your own" from the start. You just state your needs, and it will search the web, extract key points, and generate structured content.

Dia's strength lies in its cross - web page collaboration ability. You can open five long videos, several training plans, and three Xiaohongshu posts simultaneously, and then ask Dia, "I'm 170 cm tall, weigh 120 kg, love playing football, and want to focus on training my glutes and legs. Based on these materials, create a 45 - minute daily strength training plan for me."

Let an AI browser watch videos for me | Image source: Dia

Genspark is from China and can generate "the three main Office applications" based on web page information.

Comet is also an ambitious dark horse. I used Comet to open several academic papers, a 1.5 - hour lecture video, and a Google Docs document simultaneously. I then instructed it to "extract the key paragraphs about 'deep learning imitating the human attention mechanism' from the papers, write a 300 - word summary based on the professor's lecture content, and finally format it in my Google Docs."

Comet automatically completed the entire process in the background without my intervention and without disturbing me from doing other things.

Use Comet to create a unique Citywalk route according to my needs | Image source: Comet

"This is the only way to build an end - to - end workflow (allowing users to complete a series of continuous tasks in a unified environment)," said Comet CEO Srinivas.

These browsers have different styles but similar underlying logics. They have overcome the limitation of traditional browsers, which are just for accessing information, and have aggregated isolated web page information.

The interaction logic has also changed. Now, humans interact with AI, and AI operates the browser to execute commands.

Welcome to the Era of "Command - Based Web Browsing"

Recently, when I was testing Atlas, I found that it has further improved the interaction between humans and AI.

Atlas uses large - language models to replace traditional search engines in answering queries and uses agents to automatically execute user commands. In these aspects, Atlas is similar to other AI browsers.

"Memory" is one of Atlas's core selling points | Image source: ChatGPT

Let me give you an example.

Some time ago, I encountered a major bug while playing a game, which bothered me for several days. On Dia, I opened numerous forums, posts, and YouTube videos, and it analyzed them to summarize the cause of the bug and the solutions. This was already quite intelligent.

However, on Atlas, in addition to that, it remembered my earlier mentions of computer configuration, display settings, the posts I had read, and even the methods I had tried. Then it generated a targeted suggestion: "This bug might be related to the DLSS mode. You mentioned that the game was set to 'ultra - high quality.' I suggest trying DLAA or turning off ray - tracing to test the frame rate changes."

For the same command, Atlas's response above links all the posts I've read recently and will actively search for more. Dia below doesn't go beyond the currently opened web pages | Image source: Self - made by the author

Atlas remembers my previous questions | Image source: Atlas

Another example is when I spent several days planning a trip, searching for travel routes, checking flight tickets, and browsing hotels. When I thought I had seen enough, I told Atlas, "Organize the hotels I looked at last time and recommend a few more of the same type. Rank them by location and price."

Then Atlas retrieved my past conversations and web page records, determined the location, price range, star rating, etc., of my target hotels, found more hotels of the same category, and summarized the pros and cons of different hotels for me.

This is the ability to "understand users." The reason is that ChatGPT is the "core" of the browser.

Atlas is a strategic signal from OpenAI. All your needs can be directed to ChatGPT, and then AI will help you search the web for answers, whether you want to query information, create content, or work...

"We believe that AI represents a once - in - a - decade opportunity to rethink the browser," Sam Altman said when launching ChatGPT Atlas.

In the past, browsers existed based on web pages. This is related to the original design of the World Wide Web. However, AI is reshaping the way we acquire and process information, and now we are no longer "revolving around web pages."

We don't need to operate web pages ourselves and search for answers laboriously. Instead, we can issue commands and let AI execute them on our behalf.

Comet CEO Srinivas once said that he wanted to turn the browser into an operating system. This is the ambition of all AI browsers. The AI agent is the core of the system and will complete all tasks assigned by users, enabling the browser to change from "browsing" to "doing things."

Every paradigm shift is accompanied by a "big discussion" about trust and ethics, including privacy, data boundaries, and information control... This has always been the case in the history of technology. Moreover, it's even more difficult to change users' habits.

Well, Atlas gives me a feeling similar to that of the first - generation iPhone. It may not have the most comprehensive functions, but it is the most like the future.

Cover image source: Atlas

This article is from the WeChat official account "Guokr" (ID: Guokr42). Author: Gao Ji Dong Wu. Republished by 36Kr with permission.