Welcome to the web browser's next chapter
Comet, Neon, Dia, and Mariner aren't just experiments. They're blueprints for a new kind of web interface powered by agentic AI.
Sure, Perplexity's just-announced web browser, Comet, might feel niche. It's invite-only, tied to a $200 monthly subscription, and lacks some of the familiar polish and plugin support that most of us expect from our daily drivers.
Before I begin, I should note that I received an early invitation to Comet and have started testing it. It’s built on Chromium, which means it inherits much of the performance and compatibility users expect: snappy load times, full support for Chrome extensions, and seamless import of browsing data like bookmarks and history.
That foundation lowers the barrier to trying something new, especially when the broader interface shift is as significant as what Comet is aiming for. I’ll have more to share once I’ve spent more time with it, but it’s off to a solid start.
But what makes Comet worth watching isn't the product itself. It's what it represents.
Comet is built around the idea of the agentic browser. It doesn't just show you information, it helps you act on it. Summarize an article. Compare five tabs. Plan a trip. Write a follow-up email based on what you just read. It's not a search box bolted onto a tab; it's a context-aware assistant that lives alongside your session and remembers what you're doing.
And it's not alone.
This May, Opera launched Neon, calling it the first truly agentic browser. Neon doesn't just assist, Opera says. It initiates. With dedicated "Chat," "Do," and "Make" modes, it's designed to interpret intent and execute it, even if the user never explicitly issues a command. It can automate tasks, generate content, and keep working offline via distributed cloud agents. It's Opera's attempt to completely reimagine the browser, not just as a viewer of the web, but a participant in it.
Then there's Dia, a surprising pivot from the team behind the web browser Arc (which I loved, by the way). The Browser Company recently announced that it is shifting its focus away from Arc and toward Dia as its new AI-first browser, designed around command and conversation. The address bar becomes a control center. You can tell Dia to write, summarize, find files, send emails, or complete tasks directly from the interface. It's not that Arc failed, the company says, but that Dia represents where the interface is heading.
Meanwhile, Google's Project Mariner is a research-driven effort that's starting to integrate into Gemini and Chrome. Mariner positions the browser as a service interface: a place where AI agents can browse, shop, submit forms, and complete multi-step workflows autonomously. If earlier AI features inside Chrome were about enhancing the user's behavior, Mariner would be more about replacing them to free the user up to do other things.
Together, these moves by tech companies mark not just a trend, but they reveal a larger battle that's starting to unfold. Not just over features, but over who controls the interface layer of the internet.
The fight for interface (and the power it brings)
In my previous post, I discussed how the Model Context Protocol (MCP) and LLM platforms are increasingly abstracting the web from users. Content is being pulled into AI systems, reformatted, summarized, and served up without attribution or direct interaction. In that world, the open web becomes the backend infrastructure for a new kind of front-end: the chat interface, the assistant, the agent.
Agentic browsers, such as Comet, represent the next phase of this shift. But instead of replacing the web with something else, they're redefining the entry point to it.
If the old web was driven by user intent — where you search, click, read, and decide — the new web is shaped by agentic suggestion. This term refers to the browser's proactive role in guiding user actions. The browser doesn't just wait for you to act. It helps choose the action.
The browser may recommend steps, pre-fetch content, generate a draft, or organize your tasks without being asked. The user is still involved, but not always in the driver's seat. And that changes the power dynamics of the internet in subtle but profound ways.
The more your browser anticipates your behavior, the more you're relying on its reasoning. And if that reasoning is shaped by product design, model training, corporate priorities, or monetization strategies, then your sense of choice becomes increasingly mediated. Who owns that interface? Who designs the defaults? Who benefits from nudging your attention, your workflow, and your purchases?
The answer to those questions is rapidly becoming as important as which browser is the fastest or which one has the best tab grouping. Because the interface is no longer neutral. It's becoming a place of influence and possibly control.
A reordering of how the web works
Some of these agentic features are genuinely useful. As someone who uses all the latest LLMs (including the pro versions of ChatGPT, Perplexity, and Gemini), I'm excited to further explore these new browsers and their features.
Comet, Dia, and Neon all hint at a smoother, more integrated browsing experience where tab chaos and multi-step tedium get handled in the background. If the previous wave of browser innovation was about privacy and speed, then this one is about workflow fluidity.
Browsers are becoming multi-tool environments. They are places to create, research, organize, and execute without switching apps or copying links. This evolution promises a more efficient and productive web experience for users. However, there's another layer here as well.
As AI systems increasingly treat the web as an API layer — a set of resources to be queried, parsed, and acted upon — the browser becomes more than just a surface-level product. It becomes the governor of that interaction. It decides what gets passed to the agent, what gets remembered, and what gets buried.
We're witnessing a quiet migration: from humans interacting with websites to agents interacting with data and users simply approving or tweaking the results. Should it become the new normal, this shift has implications for the nature of the web, the role of users, and the power dynamics of online interactions. The open, link-based web is being folded into systems that act on our behalf, whether we see how the sausage is made or not.
Before the defaults harden into place
Over the years, I've cycled through Vivaldi, Brave, Tor, DuckDuckGo, Arc, and numerous other browsers — all with their own philosophies about what the web should feel like. But this moment feels different. This isn't about tab management, ad blocking, or dark mode. It's about shifting where control lives. Browsers like Comet, Dia, and Neon aren't just user tools. They're active mediators, and eventually, they may become agents in their own right.
The decisions being made now about memory, agency, transparency, and extensibility will shape the norms of the next decade of online life. Just like how one company ended up winning the search interface, someone will no doubt win the agentic browser interface. And whoever does so may not just shape the web, but they could also become its new gatekeeper.
This is the time to closely monitor how these browsers evolve before the new defaults become established. Before agentic behavior becomes assumed and browsers are no longer simply windows into the web.
Because if browsers are becoming co-pilots instead of librarians, maybe we should be asking: co-pilots for whom?