A $200 AI browser wants to teach me how to "surf the Internet" all over again.
As the battle of AI browsers rages on to this day, from Arc's early attempt to reshape interaction, to Opera Neon's demonstration of "proxy" capabilities, and then to the rumored browser to be launched by OpenAI, every major player in the industry is trying to redefine this most familiar internet gateway for us.
This week, Perplexity, known for its AI search engine, finally entered the arena with their answer - Comet, a browser self - proclaimed as "AI Agent native".
However, Comet hasn't generated enough buzz on social media because it is currently only available to Perplexity Max subscribers and users with certain limited invitation codes (the monthly subscription fee is $200). Subsequently, the user base will be gradually expanded through a waitlist.
Fortunately, GeekPark was able to quickly experience Perplexity's AI Agent browser through an invitation code.
Aravind Srinivas, the CEO of Perplexity, has ambitious visions for Comet: "We built Comet to enable the internet to do what it has always yearned to do: amplify our intelligence." The core concept of Comet is "From Browse to thinking".
It sounds grand, but what exactly can Comet bring to the existing browser experience? How is it different from the Dia browser with integrated AI capabilities or Chrome, which will soon have Gemini?
Can Perplexity, which started with AI search, push its valuation to new heights with Comet?
What is a "thinking partner"
To understand Comet's ambition and its current "longest suit", we still need to start from "what are the user needs of an AI Agent browser in Perplexity's eyes".
If traditional browsers solve the problem of "accessing" information, then Comet tries to solve the problems of "understanding" and "applying" information. It believes that the root of the problem is that each tab is an information island. Its solution is to connect these islands into a continent with unified intelligence.
This concept is reflected in every aspect of Comet; it doesn't quite resemble a traditional browser homepage. It's more like the desktop of a smartphone, with various apps you need arranged on it.
Comet browser desktop | Image source: GeekPark
A traditional browser is like a huge building composed of countless independent rooms (tabs), each storing different information. You need to visit each room in person to collect and organize. Comet, on the other hand, tries to transform this building into an intelligent entity with a unified central nervous system. You just need to stand in the hall (Comet Assistant) and give instructions, and this intelligent entity will visit all the rooms for you and bring back everything you want. This is a paradigm shift from "space management" to "intelligent delegation".
The core weapon for Comet to achieve its grand vision is the Comet Assistant located in the sidebar. Its magic comes from the deep integration of two major capabilities: one is "context awareness" beyond a single page, and the other is "proxy execution", which can turn information into actions, similar to what we've seen in Manus AI before.
This experience is different from previous AI browsers that rely on single - page information reading to achieve AI effects. Comet's ability has the potential to further change the way we handle complex information flows.
Imagine you're doing research for buying a new camera. You have several tabs open in your browser: product pages on e - commerce websites, in - depth reviews on professional photography websites, hands - on videos on YouTube, a blog post comparing it with competitors, and a forum discussion about its drawbacks. In the traditional workflow, this is bound to be a tough battle of constantly jumping between different pages and using your brain or notebook to record and compare.
But in Comet, this process is completely re - engineered.
You can directly ask the assistant: "Based on these open pages, comprehensively summarize the advantages and disadvantages of this camera. In particular, how does it differ from another competitor in terms of video functions and operability? Present the results in a table. Also, what do professional review websites think of the low - light image quality problem that users complained about in that forum post?"
At this time, Comet Assistant plays the role of a top - notch professional assistant | Image source: GeekPark
It can quickly read and understand the content of all pages, including video subtitles and forum discussions, and then generate a well - structured, in - depth report that synthesizes various viewpoints. This is the power of "context awareness", which integrates isolated tabs into a unified and dynamic Browse Session, and this session is its memory and workspace.
I no longer need to browse myself, but let my intelligent Agent do it | Image source: GeekPark
This ability is not limited to consumer research. It really shines in more complex professional knowledge research work.
Suppose you're writing a market analysis report, and you have a PDF industry research, a Google Sheet data table, and a draft of your Google Docs report in your tabs. You can directly give Comet a series of continuous Agent instructions: "Extract all the key data about market size and growth rate from the third chapter of that PDF, then fill them into the open Google Docs manuscript, and generate three core strategic suggestions."
In this series of continuous commands, after generating the corresponding content, Comet Assistant can fill this online document in the correct layout in an AI Agent way.
Comet Assistant can read information from multiple web pages and operate simultaneously | Image source: GeekPark
Of course, you can also make further requests to it: fine - tune the format, further enrich the details, or even let it directly come up with a title and modify it automatically.
Comet can complete more complex task requests by simultaneously monitoring and operating multiple web pages | Image source: GeekPark
To provide a more seamless experience, Comet will also ask you for permissions to read your calendar and emails to offer more personalized Agent assistance services.
Comet asks users for various permissions at the beginning of use | Image source: GeekPark
In addition, AI Agent is also an important capability that Perplexity has added to Comet. Comet allows AI agents to directly execute tasks in the local browser (such as batch web page operations, automated forms, cross - platform operations, etc.) without relying on a cloud virtual environment. The process is smooth and there's no need for repeated logins.
You can directly present your needs to it, and Comet can automatically understand and open the corresponding website to help you modify your personal information. This is why Comet asks for many sensitive account permissions at the beginning - but you don't have to worry about the risk of data leakage because these web page information editing operations are based on local processing.
Here, Comet has gone beyond being an information integrator and has become an executor of workflows. It not only helps you "see" but also helps you "do".
The biggest selling point of Perplexity Comet is that it truly achieves browser - level automation and deep AI integration, making "letting AI truly surf the internet and do things for you" a practical scenario for a new generation of productivity tools.
The "strategic choices" of AI browsers
In terms of actual experience, Comet can be said to be one of the most well - rounded AI Agent browsers at present. It's also the second browser after Arc that makes me think about "switching my default browser from Chrome". But does this really mean that Comet can survive in the wave of AI browsers?
Facing the AI wave, browser products on the market have actually chosen three distinct evolutionary paths. Comet's choice determines its unique positioning and also foreshadows the challenges it will face.
The most common and conservative path can be called the "tool enhancement school". Represented by Chrome with integrated Gemini and Edge with integrated Copilot, their core logic is "browser + AI". AI is integrated as a powerful new function, allowing you to more conveniently summarize web pages and polish text. This is useful, but the basic form of the browser and users' usage habits remain unchanged. AI is just a more useful new tool.
The implementation of Gemini in Chrome that we see today is a well - known representative of this school | Image source: GeekPark
Going a step further is the "proxy execution school". Represented by some exploratory projects, they enable AI to more actively operate the browser according to users' vague intentions, and even generate reports or applications for users in the cloud. Here, the role of AI has been upgraded from a "tool" to a "junior assistant" with a certain degree of autonomy.
What Comet has chosen is the third, and most radical and imaginative path - the "environment reconstruction school". Explorers on this path believe that in the AI era, AI should not just be a function of the browser; the browser itself should be an AI environment. Their goal is to completely redefine the form of the browser, unifying the fragmented web page information flows into a continuous, conversational, and intelligent interactive environment.
Perplexity believes that as people increasingly use AI chatbots to obtain information, traditional search and browsing patterns are changing. Comet aims to seize this trend and attract users by providing a more efficient and intelligent AI - driven experience.
Therefore, Comet firmly chooses to be part of the "environment reconstruction school", which means that its expectation for users is not just to "use" it, but to "inhabit" in it. It hopes that users will change the way they use the internet and regard the browser as a "thinking partner" that they can have in - depth conversations with and entrust tasks to completely, rather than just a passive window for displaying information.
High entry fees and user "inertia"
However, choosing the most radical path also means facing the steepest cliff.
Comet's release has not been smooth sailing. Its strategy and concept have put it in the typical dilemma of innovators.
First, there is the highly controversial release strategy. Currently, the experience qualification for Comet is only available to Perplexity Max subscribers who pay as much as $200 per month. This has greatly disappointed a large number of Pro users who pay $20 per month and are its core supporters, making them feel "betrayed".
A user's comment on social media represents the feelings of many: "It's a complete emotional roller - coaster... We thought Pro would be next." Although Perplexity officially promised that Comet will eventually be free for all users in the future, this "$200 entry fee" has undoubtedly labeled it as "elitist" and "out of touch with the masses" in the early stage, greatly limiting its current word - of - mouth spread and the establishment of an early - stage user ecosystem.
Many users have expressed their anger at Comet's current testing strategy | Image source: Twitter
Deeper than the price controversy is the huge challenge regarding user habits. When The Browser Company reflected on its widely - praised Arc browser, it frankly admitted that the core reason why Arc was cool but failed to achieve large - scale popularization was that it "was too different, there were too many new things to learn, and the rewards were too few".
This is the "Arc lesson" that terrifies everyone in the AI browser circle - it accurately points out the core contradiction faced by all "reconstruction school" products: If it's too conservative, users have no reason to abandon the mature Chrome ecosystem; if it's too radical, users may give up before truly experiencing its value due to the high learning cost.
Comet is the embodiment of this contradiction. The "conversational" browsing experience it offers, although it may mean an exponential leap in efficiency for some users, is like asking most users who are used to Ctrl+T (open a new tab), Ctrl+W (close a tab), and jumping between tabs to learn a brand - new "language". Comet must prove with undeniable value far beyond existing tools that this