Original video in Chinese.
Key Takeaway
- AI browser positioning: a web agent for ordinary people, meeting needs like intelligent search/summaries, automated tasks, and multimodal interaction; Perplexity Comet is the best one, but you need an invite code.
- Comet capability demo: handling YouTube/Douyin videos (locating, summarizing, generating scripts); crawling blogs/sites to extract content; automation such as watching for uploads and clicking publish, filling in promo codes.
- Market potential: practical AI usage (leave a webpage open and let it handle things automatically); Perplexity is strong in engineering/product, and the AI browser may become the first killer app, with fierce competition in the space.
If Claude Code is an agent for programmers, then the AI browser is an agent for all ordinary people.
What do ordinary people interact with the most in daily life? Webpages. When it comes to webpages, what do ordinary people want from AI? In the end, it comes down to just three things: intelligent search and summarization, automated tasks, and multimodal interaction.
That’s what AI browsers are targeting. They are AI agents for the masses, an ordinary person’s “universal assistant.”
Right now, the AI browser space is already getting crowded, with several products competing. If you can get an invite code, definitely try Perplexity Comet. I shared the two invite links I generated in Knowledge Planet newtype.
I think Comet is probably the best AI browser at the moment. Its operation of browsers and webpages is already quite mature.
Look, I asked it to help me open a YouTube video, and the content happened to mention Technology is not enough. This segment is the source of the famous “standing at the crossroads of technology and humanity” line.
Comet first searched, then fetched the full content and located the relevant part. In the end, it not only opened the video, but started playback from the corresponding point in the content.
Using the same approach, when watching a video, you can ask Comet to help summarize it. Then you can pick one of the points and have Comet jump directly to that part.
What I want to tell you through these two examples is that Comet’s extraction and recognition of all kinds of content on webpages is already quite mature. Besides foreign sites like YouTube, it can also handle links from domestic video platforms, such as Douyin.
I opened Douyin in the browser, went to my own account page, and gave Comet the web link for one of my videos, asking it to summarize it. As you can see, it handled it perfectly.
If you’re in the content-creation business, you definitely know what this means. For example, find a benchmark video and directly let Comet generate a short video script under one minute. It even understands concepts like hooks. If you refine the prompt properly, the output quality will be even higher.
If Comet can handle video content, then text content goes without saying.
This is my blog site. All my written content, including video scripts, notes, and so on, is published there. For an article this long, having Comet summarize it and extract the key points works quite well.
Going one step further, I can also ask Comet to crawl the entire site and sort out all the content related to agents. Comet will try all kinds of ways to understand the site structure, then extract the content and produce the final output.
I believe that scenarios based on web content are definitely the closest to ordinary users. The web is naturally rich context, and this is where agents can really show what they can do. In fact, beyond content, Comet can also identify input fields and buttons on webpages like a human and operate them.
For example, when publishing videos in the web backend of Douyin or Xiaohongshu, you have to wait until the video upload is complete before you can click the publish button. That’s really annoying. Especially for someone like me uploading 4K videos longer than 5 minutes, and uploading to multiple platforms at once, the wait really takes a while.
At times like that, you can hand the last step over to Comet to automate. You just leave it open. It will watch the webpage for you, and as soon as the upload is finished, it will click the publish button.
Besides this use case, I’ve also seen some users have Comet fill in promo codes. Comet will search the web, try every code it finds, and keep filling them in until one works.
I know some people may look down on this level of AI automation and think only something that handles everything from start to finish is truly impressive. But you have to understand: just leave the webpage open, and Comet will take care of it itself. You can go do whatever else you need to do. That is the most practical way to use AI.
To create a truly mass-market AI application, beyond having strong AI technology, you also need serious engineering capability and product experience. These just happen to be Perplexity’s strengths. After AI search ran into a strong counterattack from Google, the AI browser is their next lifeline.
Right now, each company’s product is still quite restrained, and all of them require invite codes. Once they hit some tipping point and everything opens up, the AI browser will definitely become the first killer AI app.
OK, that’s it for this episode. If you want to learn about AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!