10:44
You've probably seen the headlines. OpenAI launched ChatGPT Atlas, the AI browser. Sam Altman called it a "once-a-decade opportunity." And 292 days later… they shut it down. Yeah. But that's just the start. AI browsers popped up everywhere, and most of them are dying. Microsoft's Copilot Mode in Edge was retired after 289 days. None of them seem to hold a candle to regular browsers. No one is using AI browsers. Why? Why can't anyone, not even OpenAI, get you to switch? Between May and October 2025, five companies launched or announced AI browsers within just six months, each pitching an agent that would browse the web for you.
One of which is Dia by The Browser Company. If you're into productivity, you've probably heard of them, they made another popular browser: Arc. It even drew investors like LinkedIn's Jeff Weiner and Figma's Dylan Field, and raised $128 million. By March 2024, it was valued at $550 million. So, what's this Dia browser? Its search bar is also a chatbot, it has a sidebar assistant that can read every open tab, and has "skills," which are saved prompts which function like automation. That last function is honestly kinda cool! And soon, everyone seemed to be jumping on the same trend. Opera launched Neon, a $19.99-a-month browser with an agent "Neon Do" which fills in forms, compares products,
orders groceries, that sort of thing. Like Dia, it also has "Cards" which are reusable prompts you can stack, plus some other features. Then Perplexity, the AI search engine research tool, launched Comet, and Microsoft started turning Edge into a Copilot-powered assistant. "Edge doesn't just wait idly for you to click but anticipates what you might want to do next." Interesting, and you know who else jumped on this? Google. But I'm not talking about Gemini. Google's browser agent, Project Mariner, was shown onstage at I/O 2025, on a $250-a-month AI Ultra plan. An AI agent, but built for using browsers specifically.
A few months later, OpenAI joined the race with ChatGPT Atlas. It had ChatGPT built into every page, could remember the sites you visited, and could even carry out tasks for you through its agent mode. At launch, it was only available on Mac. To Sam Altman, that was the time for chatbots to replace URL bars. And on Atlas's launch day, Alphabet, Google's parent company, fell 2%. But there's still a question. Browsers adding AI features I get, but why are AI companies obsessed with browsers? Well, think about it. The browser is where people spend hours every day.
Habits go pretty deep. If you can capture a layer where users naturally congregate, it can be very lucrative. For advertising, for subscriptions, and for something else. Data. In April 2025, Perplexity's CEO explained that "We want to get data even outside the app to better understand you. We plan to use all the context to build a better user profile and, maybe, you know, through our Discover feed we could show some ads there." Browsers are a goldmine of data. Preferences, habits, behavioral information, purchasing information. Imagine what your AI might know about you, just by watching you browse?
So, the recent shift in the direction of the internet has given many companies a chance to get into the game. With, at least on the surface, a compelling unique selling point. This was a "rare, once-a-decade opportunity to rethink what a browser can be about and how to use one." There's something hugely ironic about all this, which we'll get to later. But first… what happened to all of these AI browsers? Where did they go? By May 2026, Microsoft had retired Copilot Mode in Edge, leaving its browsing agent available in a more limited form to Microsoft 365 Premium subscribers in the US. But even Microsoft's own executives were questioning where Copilot belonged.
"It's critical that we remove Copilot from places where it doesn't live up to its promise." Google's attempt didn't fare much better. Project Mariner was slow, expensive, and prone to mistakes. By May 2026, Google had shut it down, moving its technology into other products. The most telling, though, was Atlas. By March 2026, OpenAI was already warning its staff not to get distracted by side quests, calling Anthropic's lead in enterprise AI a wake-up call. And Atlas was one of those side quests. In July, OpenAI announced that the browser would stop working on August 9, 2026, less than a year after its launch. Atlas had never even made it beyond Mac. And now, instead of competing with Chrome, OpenAI is folding
Atlas features into existing products from browsing inside the ChatGPT desktop app to cloud browsing for its workplace agents. Ironically, it is now a Chrome extension. But, did you know Atlas wasn't even OpenAI's first try at this? That was Operator, launched in January 2025. It could click through websites to fill out forms, order groceries, and book appointments. But just seven months later, it ended up like Atlas as a ChatGPT agent. What's also funny is that in May 2026, AI-visibility firm presenc.ai called Atlas "the breakout winner" of the AI browser race, with an estimated 10-15 million monthly users. Only for it to shut down a few months later.
There were a couple of winners though. Or maybe "survivors" is the better term. Opera Neon is still around, but, Opera hasn't mentioned it in a while. Even in their full earnings call, which included AI use in their existing browsers… No mention of Neon. Dia is still around, but is changing focus. And this is a weird one. After introducing Dia, The Browser Company also tried to bring all kinds of AI features into Arc as well. But that became a problem. "For most people, Arc was simply too different, with too many new things to learn, for too little reward." He also admitted that Arc "lacked cohesion, in both its core features and core value."
I don't think people agree, as a lot of people loved Arc, but it definitely lacked cohesion after a bunch of AI features were added. Eventually, the company began pushing users toward Dia, and ended development on Arc. Then Atlassian bought The Browser Company for $610 million, hoping to turn Dia browser into a tool that could do more than just browse the web. In September 2026, Atlassian CEO Mike Cannon-Brookes said: "We're building a doer, not a browser." Dia still hadn't released its Windows version. Meanwhile, Comet kept going. After its launch in July 2025, it went free by October, arrived on iOS in March 2026, and was still receiving updates in September. In August, it also won its appeal against Amazon.
So why did Perplexity keep pushing while OpenAI pulled back? Perplexity tried to get its AI assistant onto phones as the default, but Google's existing agreements stood in the way. So, building a browser offered another route. Unlike Atlas, Comet is built around Perplexity's own AI search engine. That means people who already use Perplexity have a reason to switch. Ironically, Comet defaults to Google for search on iOS. Even Perplexity's CEO admits that Google handles navigational searches better than anyone. Perplexity Comet has 10 million monthly active users. That sounds like a lot.
Does that prove that AI browsers are a success? Well, let's take a step back. Firefox, the fourth-largest browser, with just 2.8% of the market, had 198 million monthly active users. Almost 20 times that of Comet. And Chrome even bigger. But you know what's stranger? Perplexity tried to buy Chrome from Google. For $34.5 billion. All cash too. Yeah. Double Perplexity's valuation, by the way. It turns out this is actually an important piece of this story, and related to a bigger question: Why does nobody seem to use AI browsers? It seems they even die, or get reworked into
something other than a dedicated browser. Why have so many shut down or failed to get user adoption? If you've watched this far, consider subscribing! It helps the channel and means you won't miss future videos. There are a few reasons. First, the reality is, most people stick with whatever comes on their phone or computer. Chrome controls about 69% of the global browser market, while Safari has nearly 16%. Even OpenAI, with 800 million weekly users, struggled to convert a fraction of its users to Atlas. In fact, according to reports,
AI browsers combined account for less than 1% of the global market. Less than 1%, combined. Most people just use what's already there. Installing a new, unfamiliar browser with lots of new functions is a big ask for a lot of people. But that's just the beginning. The few users they manage to pull are put at risk. AI browser agents can read websites, click buttons, fill out forms, and take actions for you. That also makes them vulnerable to a trick called prompt injection, where hidden instructions on a webpage can manipulate what an AI does. They might be the same color as the page background, for example. You can't see them, but the AI can. People can use this to trick your LLM into sharing your private details,
sending you malware, or even connecting to remote plugins that run malicious code. Security researchers demonstrated the problem on Atlas using just a few hidden words in a Google Doc. Even OpenAI admits that: "Prompt injection, much like scams and social engineering on the web, is unlikely to ever be fully 'solved.'" Gartner even advised companies to block AI browsers until safer versions were ready. Major websites also wanted them out, too. In November 2025, Amazon sued Perplexity over Comet's shopping agent, alleging that it disguised itself as a regular browser while shopping through users' accounts. A judge temporarily ruled against
Comet in March 2026. But in August, an appeals court overturned that order, finding that the user, rather than Perplexity, was the one accessing Amazon. The legal fight isn't over, but perhaps the biggest reason is that… AI didn't really need a new browser. These AI functions are likely very helpful for some people. But, for most people, even AI users, just unnecessary. If you ask ChatGPT or Claude to "search for you"… it just does it. How is putting it in a browser any better? Maybe for niche uses, like the Perplexity search engine, but for the vast, vast majority of people, that's more than enough.
Microsoft moved Copilot's browsing features into Edge. OpenAI started moving Atlas's features into ChatGPT and a Chrome extension. The extra value of using a separate AI browser, instead of using AI in a browser you already have, just wasn't that high. There's one big AI browser that I intentionally left out of this. The one that did succeed was… Google Chrome. Google started adding Gemini to Chrome, then launched Auto Browse in January 2026. It could click, scroll, and fill out forms for you, although you still had to confirm certain actions, like purchases and posts. Google also expanded Gemini in Chrome to US Android users and introduced agents that could operate Chrome using your saved passwords.
There's also the Google AI Overviews, but those are present on every browser, not just Chrome. But Chrome won this race… because it's where users already are. So it wins mostly, by default. This reminds me of the 9x effect, from Harvard Business School professor John Gourville's 2006 article "Eager Sellers and Stony Buyers". He proposes users overvalue what they already have by roughly 3x. Companies, however, also overvalue their new product by 3x. So a new product has to be about 9x better before people will switch. AI browsers can't win, by being just a little better. They have to be a way better experience.
But, do you remember how Perplexity wanted to buy Chrome? Isn't that weird? Well, it's related to a long legal battle and a judge's ruling that Google had illegally maintained its search monopoly. This is what I meant by people using what's default on the device. Google has been paying tens of billions a year, to multiple browsers, to have Google as the default. Including Safari and Firefox. Courts were actually looking to force Google to sell Chrome at one point… but they didn't. And, you'll never guess the reason. In an unexpected turn, the rise of AI browsers became one reason the judge allowed Google to keep Chrome.
Judge Amit Mehta let Google keep Chrome in September 2025 because generative AI had "changed the course of this case" and AI companies looked newly able to compete. He admitted he was being asked to "gaze into a crystal ball and look to the future." But within a year of that ruling… Atlas was gone and Chrome was the AI browser. In the end, the companies that set out to take on Chrome may have helped Google become even harder to beat. Ironic. But one of the most baffling of all these examples is Copilot. And, things aren't going well for their non-browser side either. Microsoft pushed this thing into basically every part of Windows they could, and it backfired. Not only did users find it annoying,
but it couldn't even do the most basic tasks, and would just tell you to do them. To learn more, watch this video next.
