Dia, a preview of the future of AI browsers

Over the past three decades, the evolution of browsers has been more like a slow process of tinkering. We have evolved from Mosaic to Netscape, and then to Chrome (I don’t want to mention IE at this moment). Although the interface is more beautiful and the speed is faster, in essence, they are still tools centered around “web documents”. You enter the URL or search for keywords, click on the link, wait for it to load, and look for fragments of information like flipping through a book in a library. This way of interaction has not been truly subverted to this day.

Until Dia came along.

This AI browser created by Arc is not an "AI plug-in version" of Chrome, nor is it the "successor" of Arc, but a completely new way of thinking: what would happen if AI was the core of the browser instead of a plug-in? Therefore, Arc had a clear goal from the beginning: to build a native AI browser from scratch  .  

From Arc to Dia: From geek tools to mass access

Arc is a browser made for geeks. Its features such as tab management, split screen, and Spaces are amazing, like a fine work of art, sharp, complex, and fascinating. But it is also because of this that it has always been difficult to break out of the circle. It requires users to have a certain system thinking and operating habits, like a Swiss Army knife that requires an instruction manual. Josh Miller, the founder of Arc, once said frankly: "It is too complicated and too different, and can only be a tool for geeks."

The emergence of Dia is a transition from "tool" to "entry". It does not continue the vertical tab bar of Arc, but returns to the familiar horizontal tab page; the homepage is so simple that there is only one input box, but it hides a powerful AI engine. This "restraint" is not a compromise, but a philosophy: it is not to make users adapt to tools, but to make tools adapt to users.

"Entrance" is not only a UI element, but also a metaphor for cognitive mode. Arc is designed for people who "know what they want", while Dia is for people who "are still searching".

Dia's input box is no longer a "search box" in the traditional sense, but an interface to thinking. You can enter keywords to search for information as in the past, or you can directly ask questions to AI, request suggestions, and generate content. It no longer distinguishes between "search" and "conversation", but merges the two into a more natural way of interaction. You don't need to decide "I want to search" or "I want to chat", you just need to express your intention, and AI will understand the context and give a response.

This concise input box provides two options: Google Search and Chat. When receiving user input, the default is to call Google Search, continuing the usage habits of Chrome users; but when the user enters more content and a line break appears, it immediately switches to Chat mode. These details are surprisingly polished.

Behind this is a respect for the human cognitive process. Information acquisition and thinking are essentially continuous and should not be separated by tools. The search box of a traditional browser is your "entrance" to the Internet; while the input box of Dia is your "interface" to AI. It is not just a tool to open web pages, but the starting point of your thinking and expression.

AI is no longer a plug-in, but part of the operating system

Dia's real ambition is not to be a "smarter search engine", but to be the prototype of an "AI operating system". As Josh Miller said, "Adding a few AI buttons to the product interface is not a revolutionary breakthrough. Dia wants to make AI pervasive in the computing environment." It is not about adding AI to the browser, but about making AI the browser itself.

Even on Dia's official website, there is only one sentence.

• AI won't exist as an app. Or a button. It'll be an entirely new environment — built on top of a web browser.

After using it for half a month, I think this sentence is Dia's gene.

For example, its smart cursor .

According to Josh Miller, the smart cursor will provide intelligent input suggestions in all places where input is needed, such as continuation, expansion or summary. Its interaction is also very interesting. Just put the cursor in the input box, or select a piece of text, and the smart cursor will appear, quietly becoming thicker and blue. Click it (shortcut key ⌘ + E, or click the Chat button in the upper right corner of the browser), and the AI ​​panel on the right will pop up. I saw someone comment that Dia's smart cursor is too hidden and users can't find it. My view is exactly the opposite. This design will be a model of the input box in the AI ​​era. It integrates AI into the input box in a subtle way.

For example, if you want AI to polish a paragraph of an article, you only need to select the paragraph and click the smart cursor. Dia will call out the AI ​​panel, understand the context of your current page, give you modification suggestions, or even write it for you. Click  Insert to replace it with one click. You may have noticed that the smart cursor is not limited to the input box. And all this, no plug-in, no jump, no "/" command is required.

Another example is  the @Tabs  function.

As long as you enter@ Symbols can be used to feed any open web page to AI as context. It's like when you are talking to an assistant, you hand him a few books and say, "These are the ones I just read. Please refer to them." AI is no longer an isolated conversation robot, but a part of your information flow. Compared with those knowledge base products that need to be sorted, classified, and annotated, this method is much lighter and more natural. I think it is a knowledge base product that ordinary users can use quickly.

Multimodal Capabilities: Understanding More Than Just Words

Dia can not only read web pages, it can also "watch" videos and "listen" to podcasts. When you open a podcast on YouTube, Dia can skip ads, extract subtitles, and even summarize the video content. I tested many videos and it responded very quickly. Even after the ad just started playing for a few seconds, Dia was able to summarize it. How does it do this? If you know, please tell me.

This is not a simple "AI summary", but an understanding and reconstruction of multimodal content. It allows you to actively extract value instead of passively consuming information.

Split-screen interaction: parallel processing of information

Dia's split-screen feature is an underestimated basic capability. It allows you to open multiple pages side by side in one window, without jumping back and forth between tabs or relying on memory to maintain context. You can view, edit, and compare different blocks of information at the same time, just like processing tasks on an orderly workbench instead of rummaging through a pile of messy drawers. Our brain is not a linear processor, but a system that constantly switches between multiple contexts. The significance of split screen is that it respects this way of cognition, allowing you to have a more natural operation path when facing complex problems.

The operation is also very simple: Control + Shift + = to add a split screen, Command + W to close the split screen. Remember these two shortcuts, and you will find that switching perspectives and organizing information becomes much smoother.

The above is a screenshot of Dia when I was writing this article. On the left is the YouTube video "An early peek at Dia, our second product | A recruiting video", in the middle is me editing this blog on the Notion website, and on the right is the Chat interface. Dia allowed me to watch, ask, think and write at the same time.

To extend this, this is not just a "product screenshot", but a screenshot of a "state of mind". You are browsing, recording, asking questions, and generating content - and these actions are no longer separate, but parallel. Dia's split-screen capability, smart cursor, context-aware AI response mechanism, and she can even chat based on two web pages are all quietly emerging in this picture. It is not "showing functions", but "showing how you think".

What’s more interesting is that this picture itself also reflects Dia’s philosophy: it’s not about making you adapt to the tool, but about making the tool adapt to you. You are not forced to jump pages, switch windows, or copy and paste, but rather flow in an organic space. Information, thinking, expression, and feedback all flow together naturally like water.

Personalized settings: AI can also "think like you"

Dia also provides an entry to “Personalize Dia”.

You can tell it who you admire, what kind of expressions you like, and what tone you prefer. It will adjust its answer style accordingly, and even build a "personality projection" that is similar to your values. AI is no longer a cold tool, but a partner with whom you have a tacit understanding.

I have set up these people: Steve Jobs, Elon Musk, Patrick Collison, Paul Graham, Naval Ravikant, James Clear, Brian Chesky, Jack Dorsey, Naval Ravikant, Richard Feynman, you can refer to them.

After simply setting up Dia's style, you can feel that the Dia development team must have put a lot of effort into "making AI speak human language according to its own wishes". It is very comfortable and elegant. Most of the content of this article was created by Dia according to my settings. I admit that I don't have such good writing skills and philosophical thinking. You can feel it.

The future of browsers is not about being faster, but about understanding you better

The logic of traditional browsers is: you look for information, it displays it, while the logic of Dia is: you express your intention, it completes the task.

Behind this is a fundamental shift: from "human operating the interface" to "human expressing intentions, AI executing tasks." This is the core of Agent thinking.

In early demos of Dia, it can already automatically browse Amazon and add items to a shopping cart. In the future, it may help you book flights, write emails, summarize meetings, and even become the "second brain" of your digital life.

Last words: Is Dia the end or the beginning?

Some say Dia is just Chrome with AI integrated in. Others say it’s the browser’s iPhone moment.

I prefer to think of it as the starting point of a question: What would it look like if we redesigned the browser based on AI?

Dia has an answer. It is not perfect, but it is real. It is not the end, but a direction worth starting from.

In this era of information overload and scarce attention, we don’t need more tools, but less resistance. The significance of Dia is not how much it can do, but that it allows us to do less “meaningless things”