Originally published at: Dia Browser Debuts with Contextual AI Chat, But Arc Users Feel Left Behind - TidBITS
I have so far avoided writing anything in TidBITS about The Browser Company’s pivot away from the Arc browser that I dubbed “the most transformative app I’ve used in decades” (see “Arc Will Change the Way You Work on the Web,” 1 May 2023). Part of the reason is that I didn’t trust myself to write about it without using bad words. If you haven’t followed the saga—and if you don’t use Arc, there’s no reason you should have—here’s the summary.
From Arc to Dia
In 2023, a startup called The Browser Company of New York introduced a new Web browser, called Arc. It was built on the open source Chromium project that powers Google Chrome, Microsoft Edge, Brave, Vivaldi, and others. What set Arc apart was how The Browser Company reimagined what it meant to use a Web browser, turning it from a dull wrapper for webpages into a powerful tool that reduces much of the unnecessary friction in browsing. Features like pinned tabs in a sidebar, workspaces for organizing collections of pinned tabs, split view for viewing multiple pages side-by-side, a Little Arc window for displaying view-and-close pages, a Command-Option-C shortcut for copying a page’s URL, and many other quality-of-life improvements turned me and many others into dedicated Arc fans.
It helped that The Brower Company operated in a culture of extreme transparency, releasing notable new versions every Thursday with release notes that were both personal and professional. They often included videos of engineers in their apartments explaining the latest features, and Josh Miller, the company’s CEO, frequently shared videos discussing the company’s plans.
That transparency wasn’t accompanied by much interaction with users, though. Direct support was weak or nonexistent. Despite its social media-influenced approach to communication, The Browser Company never launched its own online forum, and its representatives rarely participated in independent forums. Nor did it particularly engage with the press; I had one brief contact with a rep there, but he soon left the company, and none of my subsequent inquiries have been answered.
From all outside indications, Arc was a success. However, that wasn’t sufficient for The Browser Company, which aimed to build a mainstream browser for everyone. While The Browser Company has never revealed Arc’s user numbers, estimates place them in the mid-hundreds of thousands, possibly reaching a million, compared to Miller’s goal of hundreds of millions of users. Since Arc is free and has no clear monetization plan, those user numbers are too low to meet the company’s investor expectations and ambitions.
In mid-2024, the company released 10 weekly podcasts outlining its thinking for Arc 2.0. The series didn’t end conclusively, but there was a sense that there was a plan for Arc 2.0. Instead, less than a month later, Miller told The Verge that The Browser Company was going to build a new browser.
But Arc wasn’t going away! Nothing could be further from the truth! Or so we were reassured. Indeed, weekly updates continued, though they were almost entirely limited to new versions of Chromium. The Browser Company’s claim now is that Arc is mature and essentially feature-complete, so there’s no need for updates. The unfixed bugs I’ve reported would beg to differ, and it seems odd that a company that was releasing new features at a breakneck pace for over a year suddenly couldn’t think of any more. (Although I’ve never used it, my understanding is that the Windows version of Arc has far more problems since it wasn’t nearly as far along in its development.)
Eventually, Miller posted a letter on the company’s blog that attempted to explain the shift from Arc to this new browser, called Dia. I can’t argue with Miller’s claims about the uptake of Arc’s features, the architectural problems Arc faced, or the foundational development kit that prevents the company from releasing Arc as open source. He’s an amiable and engaging guy, and I think I’d rather like him in person.
Nevertheless, Miller never apologizes for or explicitly acknowledges the anger and distrust caused by announcing the end of Arc in October 2024 and saying nothing about Dia until May 2025. He did a better job on the Waveform podcast, where he at least admitted that he wouldn’t have opened himself up to all the criticism unless he felt that moving to Dia was truly the right thing to do for The Browser Company (if not for Arc users).
In essence, The Browser Company cultivated a devoted user base for Arc, only to pivot away from those same users with Dia. It’s like getting dumped via a long letter explaining how amazing you are—they simply want to see other people, lots of other people—but hey, we should definitely stay friends!
Introducing Dia
Assuming you can get past feeling like you’ve been asked out again by someone who just dumped you, you now have a chance to see if The Browser Company deserves a second chance and if Dia is all it’s cracked up to be. Dia is now available for Arc users in beta; others can join the waitlist. It runs on macOS 14 Sonoma and later and requires a Mac with Apple silicon.
Unfortunately for Arc users, Dia is best described by the working title of the latest post from designer Charlie Deets, “Why would The Browser Company build a boring browser?” Dia is a boring browser that looks and works like a stripped-down Google Chrome, complete with the usability nightmare of top-mounted tabs. The design brief is to make Dia so boring that someone could switch to it at 10 AM on a Tuesday morning.
Someone, that is, who doesn’t use Arc. If you use Arc, you’ll immediately find yourself incapable of getting anything done without your pinned tabs, workspaces, and numerous other features. I already have 13 tabs open in Dia and can’t easily distinguish between them, whereas in Arc, most would be familiar pinned tabs that I could find and click at a moment’s notice.
Dia does include a few nice features from Arc. Pressing Control-Tab quickly switches you back and forth between your two most recent tabs, and continuing to hold the Control key brings up Arc’s tab switcher. Option-clicking a link or pressing Option-Return in the Command Bar opens the page in a split view. Pressing Command-Option-C copies the URL of the current page. The best part of the recent design post is where Deets says:
We want to bring more of the functionality people love from Arc to Dia. Concepts like sidebar tabs, tab management and incoming link routing for profiles are high on our priority list.
That’s a step in the right direction, but I worry it will fall short of what Arc provided in important ways. Vertical tabs are essential, but unless they’re pinnable and can be separated into workspaces, most of Arc’s productivity gains will be lost. On the Waveform podcast, Miller suggested that more Arc features would start to appear in Dia later this year.
If Dia is so much like Chrome, why would anyone bother with it?
Adding AI to the Browsing Experience
The main idea behind Dia is that browsers serve as our primary access point to the Web, but increasingly, people also want to engage with AI. In the Waveform podcast, Josh Miller says:
What is a browser? It is technically a user agent. Your browser is designed to represent you to webpages and webservers and bring stuff back on your behalf. And so it seems so clear… that people wanted to interface with the internet not just with webpages anymore, but with AI models and probably in the future, agents like deep research. And shouldn’t your interface to the Internet be able to both handle webpages and chat and models and agents?
To that end, Dia offers a chat sidebar that allows you to ask questions about the webpages you’re viewing. In fact, I used Dia to extract the above quote—complete with a link to the relevant spot—by loading the YouTube version of the podcast in a Dia tab, opening the chat sidebar, and asking, “What does Miller say about browsers as user agents?” Dia even provides Copy as Text and Copy as Image buttons for each response.
The key to Dia is its ability to grasp the context of your current page while you chat. For example, I’ve pulled up TidBITS Talk threads and quizzed Dia about their contents—no need to switch to ChatGPT and reference URLs. You can also feed Dia extra context from other open tabs, your browsing history, or bookmarks. I’ve only experimented with this so far, since my usual workflow involves opening a handful of tabs (sometimes in split view) and flipping between them as I write. Dia’s approach could make it much easier to gather and manage reference material while working.
You can also ask Dia to compare multiple tabs or compile information from them. Think about how that might be helpful if you’re reviewing a few Airbnb listings and want a table showing their prices, number of bedrooms, and proximity to a specific attraction. Or you could ask Dia questions about the listings, like which one would be best for a certain type of vacation.
Like other chatbots, Dia can handle both text and code, though I’m not sure who writes code directly in a browser tab. Personally, I spend my days wrangling text in countless contexts. When I asked Dia to analyze this article, it responded almost exactly as I’d expect from the AI-powered Lex word processor or ChatGPT. A particularly handy feature allows you to select any portion of text and ask Dia to edit, recast, or improve it. If you like the result, just click Insert to replace your original—no need to copy and paste. Unfortunately, it doesn’t track changes to show what it would modify, and as you can see in the screenshot, it heavily altered my text.
It took me a few tries to find a prompt that just did proofreading without wholesale rewriting of my words: “Identify only proofreading errors.” That explained the errors I had introduced but didn’t give me a version I could insert with a click. I eventually landed on “Identify only proofreading errors and give me a version of the paragraph that doesn’t include them, with no other text in the insertion.” (On a subsequent reading, that sounds slightly nonsensical without the context of previous prompts and responses.)
I want to remember that prompt, and Dia offers a feature for that: skills, which are basically canned prompts. When you hover over a prompt, a Save as Skill link appears. Clicking it allows you to save the prompt and recall it with a /command. In this case, I saved my proofreading prompt so I can invoke it again with /proof.
As much as I’m tremendously annoyed at The Browser Company for how it has treated Arc users, I have to admit that I think it’s on the right track. I frequently create a split view in Arc so I can use ChatGPT to research the topic I’m covering, but it doesn’t know what I’ve written. Even more to the point, I’m doing most of my writing in Lex these days, and its chatbot has full awareness of what I’ve written so far or text I’ve selected. It’s a big help for research, editing, analysis of my articles, and overcoming occasional brain lock on how to word something.
Although you wouldn’t guess it from how The Browser Company talks, Dia isn’t alone in integrating AI chat capabilities into browser sidebars. Microsoft Edge now has Copilot (shown below), Brave introduced Leo, and Opera One provides some contextual abilities through its Aria assistant. Looking ahead, Google Chrome is testing Gemini integration, and Perplexity is rolling out Comet to those on a waitlist. While I haven’t extensively tested these other browsers, Dia currently seems to offer more comprehensive integration with browsing activities, such as managing multiple tabs and inserting text.
For Arc users, Dia is mostly just a curiosity today. It may eventually incorporate enough of Arc’s compelling features to become a viable alternative, but it certainly isn’t there yet. The Browser Company tends to move quickly, so new features could arrive frequently. You may be more interested if you use a more traditional browser, are interested in AI, and can get a copy of Dia. However, you might also consider Brave, Edge, and Opera One. Notably, Apple has said nothing about adding such features to Safari, although opening its foundation models to developers could lead to some extensions that let you chat with a chatbot about the page you’re viewing.
