AI browsers need the open web. So why are they trying to kill it?

For those of us who earn a living publishing content on the open internet, Amazon’s lawsuit against AI startup Perplexity can seem darkly amusing.

Perplexity is among the many AI companies that has spent years extracting value from the internet in exchange for little. Its crawlers have synthesized endless amounts of content from publishers, even working around publishers’ attempts to block this behavior, all so Perplexity can summarize content without having to send traffic to the websites themselves.

Now Perplexity and its rivals are going a step further, with a new wave of AI browsers that can navigate pages automatically. Perplexity has Comet, OpenAI has ChatGPT Atlas, Opera has Neon, and others are on the way. The pitch is that AI “agents” will soon be able to trudge through the web on your behalf, booking your flights, buying your groceries, and shopping on sites like Amazon. Both Perplexity and OpenAI view these browsers as imperative in their goals to build AI “operating systems” that can manage your life.

Amazon, which has a lot to lose if people stop accessing its website directly, is suing to stop that from happening. It’s been trying to block Perplexity, but so far to no avail.

Therein lies the irony: These AI browsers promise a future where you’ll never have to visit a website again, yet that promise depends on having viable websites to crawl through in the first place. Amazon’s lawsuit is a sign that these two goals may be incompatible.

Feeding the beast

For companies like Perplexity and OpenAI, web browsers are suddenly important because they open the door to content and data that would otherwise be inaccessible.

Consider Amazon. If you’re just using ChatGPT’s website, you might ask it to recommend a few Amazon items or summarize a product’s user reviews, but its answers wouldn’t include any personal data from Amazon’s site. By contrast, ChatGPT Atlas and Perplexity Comet can access Amazon exactly as it appears in your own browser window. That means they can crawl through your order history or weigh in on Amazon’s personalized product recommendations.

Perplexity says these “agentic” browsers make for a better shopping experience, which is why Amazon should embrace them—but Perplexity also stands to benefit in other ways. By understanding things like your order history, personalized recommendations, and all the questions you asked Perplexity’s AI to arrive at a particular product, the company can build a much richer user profile for things like targeted advertising.

“You’ve gone from behavior tracking to psychological modeling,” says Eamonn Maguire, who leads the machine learning team at Proton. “Where you have traditional browsers tracking what you do, AI browsers infer why you do it.”

This isn’t speculation. Perplexity CEO Aravind Srinivas said on the TBPN podcast earlier this year that its browser will enable “hyper-personalized” ads by understanding more about users’ personal lives.

“What are the things you’re buying, which hotels are you going to, which restaurants are you going to, what are you spending time browsing, tells us so much more about you,” Srinivas said.

Amazon, meanwhile, has much to lose from AI shopping agents, even if they ultimately help make a purchase. The company has its own $56 billion advertising business, fueled in part by the ads it stuffs into its shopping pages. CEO Andy Jassy has acknowledged that AI agents could disrupt that business.

You may have little sympathy for Amazon in that scenario, but consider also the many smaller entities that stand to lose from an agentic web. Your favorite newsletter, for instance—one that paywalls content for its most loyal readers—may now have that content exposed within the tabs of an AI browser.

Eamonn also gives the example of research papers that sit behind paywalls, or personal documents that wouldn’t exist on the web at all. The contents of emails, shopping lists, and productivity apps could all become fodder for AI to learn more about you. And while Perplexity and OpenAI have said they won’t train AI models on what people view in their web browsers, Eamonn says they could easily change that policy in the future.

“Cynically speaking, it’s a smart way not only of building particularly good profiles of users but also getting more data,” Eamonn says.

Why the web?

Srinivas has acknowledged that AI companies need the openness of the web to provide them with all this context, because other platforms are too locked down.

“The only reason we’re doing a browser is there’s no other way to build an agent with enough control over many applications simultaneously,” Srinivas said at the Upfront Summit in February. “Especially on iOS, you cannot even access another app. You don’t want to be bottlenecked by how Apple is building its ecosystem. You want to work around it, and the browser is a very good work-around in the short term for us.”

OpenAI has similarly described the web browser as key to its broader ambitions.

“Now that we have feedback and signals from hundreds of millions of people around the world, it’s clear ChatGPT needs to become so much more than the simple chatbot it started as,” Fidji Simo, OpenAI’s CEO of Applications, wrote in a blog post announcing the Atlas browser. “Over time, we see ChatGPT evolving to become the operating system for your life: a fully connected hub that helps you manage your day and achieve your long-term goals.”

While AI companies have clear ideas of what they can do on the open web, it’s less certain whether the open web will cooperate. Lots of websites already attempt to block AI crawlers—Reddit has even cut off search engines that don’t provide compensation—but AI browsers represent yet another way around those restrictions. Amazon’s lawsuit against Perplexity could be a sign of further fights to come when attempts to block AI fail.

AI companies would have you believe that these efforts are just delaying the inevitable. But that raises a bigger question of what the open web even looks like if it becomes entirely intermediated by AI. A common complaint against AI tools like ChatGPT is that they’ll erode the incentives to create new content, and that AI itself will ultimately suffer from having nothing new to train on.

“Nothing really gets better unless you have content, but the content is getting worse because people are just using AI to generate this content, and then these models are getting worse because the content is getting worse,” Proton’s Maguire says.

With the rise of agentic AI browsers, a similar argument could be applied to the web as a whole: What motivation will exist to design beautiful, unique websites for humans when there’s no one left to browse them?

View the full article