Browsers that think for you: options, risks, and alternatives

Overnight, opening your browser is no longer just about "going online ." Now, the program you use to visit your favorite websites wants to understand what you read, summarize pages for you, fill out forms, remember your searches, and even perform tasks for you while you do something else. This is the new league of so-called AI browsers , or browsers that "think for you."

In recent months, browsers like Dia, Comet, and Atlas have emerged , while established browsers like Chrome, Edge, Firefox, Brave, and Opera are shoehorning in artificial intelligence. Meanwhile, browsers like Vivaldi, Tor, and LibreWolf are standing firm , preferring to continue offering a classic, privacy-focused experience. The result is a rather chaotic landscape where choosing the right browser has become far more important than ever before.

What does a “browser that thinks for you” really mean?

When we talk about a browser that “thinks,” we're not referring to magic, but rather to different levels of AI integration within the application itself. What a few years ago was limited to installing an extension like ChatGPT is now being integrated into the core of the browser, with three major approaches that are already becoming clear.

The first level is direct access AI . Here, the browser barely acts as a launcher: you have a box or quick access point to send queries to a language model and receive a response, and little else. The AI ​​doesn't understand much of your browsing context and functions almost like a separate website, only integrated into the interface.

The second tier is the AI ​​assistant . This is the most widespread model: a side panel with a chatbot that actually "sees" what you're reading. You can ask it to summarize the page, explain a passage, generate text from what's in front of you, or translate an article without having to copy and paste. Browsers like Edge, Brave, Firefox, Opera, and Dia operate in this area, to varying degrees; Firefox has also explored Smart Window and AI Window modes to integrate assistance more visually.

The most ambitious level is agentive navigation . This category includes Comet and Atlas's Agent Mode, where AI not only responds but also takes control of the browser : opening tabs, browsing websites, filling out forms, comparing options, and completing sequential tasks. You tell it, "Find me cheap hotels in Lisbon in June and put them into a spreadsheet," and the browser will visit several websites to do so (or at least try).

All of this opens up a range of possibilities, but also clear risks: loss of control, excessive dependence, intensive use of personal data , and new attack surfaces (such as prompt injections, malicious instructions hidden in web pages that attempt to manipulate AI). To minimize these dangers, it's advisable to review guides on key strategies for protecting your devices against digital risks.

Dia Browser: From a strange experiment to an Atlassian browser

Dia is the spiritual successor to Arc, the browser from The Browser Company that many users adopted as their daily "digital office ." Arc became famous for its vertical tabs, spacing, and almost obsessive way of helping you organize your daily tasks in front of the screen, though it did ask for something in return: learning its logic, accepting its learning curve, and thinking "like Arc." Learning to use Arc Max in Arc Browser helps you understand this AI-intensive workflow approach.

When The Browser Company announced Dia, the move was met with mixed reactions. Arc was relegated to "minimal maintenance," and the community interpreted this as abandoning it to chase the AI ​​trend. CEO Josh Miller appeared in a video talking about "something new," while users complained that they were throwing away a meticulously crafted product that still had potential.

Dia's initial steps didn't help: upon opening it, it looked like Chrome with a nicer font . No split tabs, no advanced spacing, none of the radical ideas that had made Arc special. Just a chatbot on the side and a clean, very Chromium-like interface, very familiar to anyone coming from Chrome. Many did the same: close it and go back to Arc while it was still working.

However, The Browser Company continued iterating at a remarkable weekly pace. Gradually, the features people missed began to return , often refined:

• Vertical tabs, popularized by Arc and which even Chrome has just announced it will copy.
• Visually meticulously groomed eyelash clusters, with that almost obsessive aesthetic care of the house.
• split viewThe view of multiple tabs at once in a single window, which once you try it is hard to abandon.

The Browser Company's pattern is consistent: they launch something that initially feels incomplete and even "meh ," but they polish it until it becomes addictive. With Arc, it was love at first sight on the second try. With Dia, the infatuation has been slower and more gradual, but it follows the same path.

The big difference is that Dia has decided to forgo much of Arc's quirkiness . It's much more like a souped-up Chrome: less dogmatic, less rigid, and with a virtually nonexistent learning curve. You open Dia and browse. It doesn't force you to sit through a twenty-minute tutorial on how to think in terms of tabs, something that severely hampered Arc's growth and drove away users who didn't understand its concept.

In return, some of the original's radical ideas are lost, but convenience for the average user is gained . However, Dia currently has two major drawbacks: a size approaching one gigabyte—an enormous amount for a browser—and the complete lack of a mobile version, which breaks the cross-device consistency that Arc aimed to offer.

In terms of AI, Dia falls more into the realm of traditional assistance . It features a sidebar with a chatbot that can summarize what you're reading, respond with context from multiple open tabs, or assist with small text-based tasks, but it's not the core of the experience. Many users still prefer to use Claude, Gemini, or other external chatbots for AI tasks, leaving the browser for what it has always done well: organizing tabs and workflows.

All of this is now intertwined with a major corporate change: Atlassian acquired The Browser Company in September with the aim of making Dia a leader in AI-powered work. For now, the browser's day-to-day functionality seems unchanged, but it's clear that the medium-term goal is to fully integrate it into the company's productivity ecosystem.

Comet: the agentic browser that wants to do things for you

Comet, developed by Perplexity AI , is one of the clearest examples of a browser that acts on your behalf . Its premise is simple yet unsettling: it not only assists you while browsing, but can also take control to execute complex, sequential tasks.

On the desktop, Comet presents itself as a browser that many describe as "a glimpse into the future of Chrome ." It integrates an AI model that interprets the context of what you read, anticipates your questions, and provides structured answers as you browse. If you previously typed "best camera 2025" into Google, now you could naturally ask, "What's the best option for night photography according to experts?" and Comet will search, summarize, compare, and draw conclusions using linked sources.

The articles describe it as a browser that learns from you as you browse . It analyzes your browsing history, your usual topics, your times of day, and uses that pattern to anticipate your needs. The promise is a quieter, more efficient internet with less time wasted jumping between tabs and ads.

But that same promise raises a huge question: when a tool filters, summarizes, and interprets for you, what happens to your ability to explore and think for yourself? Browsing ceases to be a process of discovery, of comparing sources and exercising critical thinking, and becomes more like accepting AI's judgment on what's worth reading.

In its most ambitious version, Comet is marketed as the first agentic browser available to the general public. It's not just a simple chatbot stuck on the side: it's an embedded agent that, if you allow it, can:

• Summarize YouTube videos in real time while you're watching them.
• fill out forms and book appointments from your email.
• Remembering research projects for days, maintaining the context.
• Perform autonomous tasks using computer vision and headless navigation.

It all sounds very impressive, but there's a catch: to work this way, Comet needs access to practically your entire digital life . Complete browsing history, email, calendar, shopping behavior, leisure activities, searches… Perplexity's CEO has openly stated that they want to know which restaurants you go to, what you buy, or how you spend your time online in order to show you more relevant ads. If you're concerned about privacy, it's worth reviewing tips on how to protect your privacy when using these types of tools.

This tension between promised privacy and hyper-personalized advertising has divided the community. On paper, data is stored locally, but things get complicated as soon as you sync with your Perplexity account. Furthermore, the browser is still experimental: slower than Chrome in normal use, with frequent crashes in agentic tasks, and an interface that many compare to a poorly designed Chrome skin.

On iOS, Comet runs into Apple's limitations: it has to use WebKit, so the difference isn't in how it renders pages, but in the automation and context layer . The browser analyzes the DOM in real time, identifies forms, buttons, and text blocks, and coordinates that analysis with language models that interpret your instructions in natural language.

All of this comes at a cost: increased CPU usage, battery drain of up to 10% during extended sessions , and a heavy reliance on cloud services, with latencies on mobile that can range from 20 to 200 ms depending on the connection. This is the price of embedding an "autonomous assistant" within the browser.

For whom does Comet make sense right now? For professionals who do a lot of research , it can be a real help, as long as it's not used with sensitive data. In corporate environments, the risk is too high. And for the average user, who wants to browse, buy a few things, and check social media, the cost to privacy doesn't outweigh the supposed magic.

Atlas: OpenAI's browser with ChatGPT at its core

If Comet was one of the pioneers, Atlas is OpenAI's major offensive in the AI ​​browser wars. We're not talking about a tab with ChatGPT open or an extension, but a browser built on Chromium where the assistant lives at the core of the product.

Atlas is initially launching for Mac, with Windows and mobile versions promised , and it comes with a powerful hook: it's free to use from day one. The goal is clear: instead of going to Google to search for something, you simply ask ChatGPT within Atlas , taking advantage of the context of the pages you have open.

The feature that raises the most eyebrows is "memory ." If you activate it, the browser remembers information about your activity: websites you visit, topics you research, usage patterns. This memory is used to give you more personalized responses. You can tell it, "Find the job postings I viewed last week and give me a summary," and the system will draw from its accumulated history.

OpenAI emphasizes that the memories are optional, private, and deletable, and that there's an incognito mode for those who don't want to leave a trace. Even so, many users feel some distrust: we've known how the internet works for years . First, it's activated for convenience, then we forget it's on, and in the end, the browser knows more about your routine than your own family.

The other key feature is Agent Mode , an experimental mode that allows ChatGPT to act within the browser almost like a person: opening tabs, browsing websites, planning trips, and filling shopping carts. In the demonstrations, the assistant read a recipe, searched for the ingredients in an online store, and added them to the cart so you only had to confirm the order.

OpenAI claims there are visible controls to stop the agent , that the system blocks sensitive pages, and that it cannot execute random code or install software. But experts have already raised concerns about prompt injections: hidden instructions in web pages capable of diverting the agent's behavior. This opens a huge door to a new type of attack, and it's only a matter of time before we see real-world cases.

Beyond its spectacular features, Atlas aims to change the usage model: to move you from navigating manually to supervising what the AI ​​does . You can request summaries of long documents, have it translate directly on the page, extract tabulated data, or automate repetitive processes without you having to click repeatedly.

This move aligns with a very clear ambition: OpenAI doesn't want to be just a provider of AI models; it wants to make ChatGPT the main interface of your digital life . First came the web chatbot, then the apps, now the browser. The next logical step is an operating system or an integrated environment where AI feels at the heart of everything you do.

The great classic browsers and their relationship with AI

While new players are trying to reinvent the web experience, established browsers are moving at different speeds. Edge, Firefox, Brave, Opera, and Chrome are all testing different approaches to integrating AI without completely disrupting the familiar experience.

Microsoft Edge was one of the first to make the serious leap. It started with Bing Chat and now relies on Copilot, which is integrated into a side panel with access to OpenAI models. From there you can:

• Chat with AI while you browse and use tab content as context.
• Use voice mode to talk to Copilot in assistant style.
• Activate writing tools that help extend, summarize or reformat texts in form fields.
• Select text on a page or PDF and click on “Ask Copilot”.

Many of these features have also been brought to Edge's mobile apps , something that many other browsers have not yet matched in terms of integration.

In the case of Mozilla Firefox , the push for AI has generated considerable controversy among its most loyal users. Some versions have suffered performance issues related to AI processes, and the more purist community doesn't look favorably on this trend. Even so, Firefox has incorporated a sidebar from which to choose different chatbots (ChatGPT, Gemini, Le Chat de Mistral, etc.) to summarize pages, answer questions, or improve texts; if you prefer to maintain control, you can disable the AI ​​features in Firefox and adjust the integration to your liking.

Brave is betting on AI with Leo, an integrated assistant that interacts directly with page content: it can summarize articles, analyze documents, or generate new text. The key here isn't so much what it does—similar to other sidebars—but the total emphasis on privacy : they promise that conversations are not used to train models, are not stored, and are processed anonymously.

Leo is available on both desktop and mobile and is perceived as a lightweight integration, consistent with Brave's philosophy of blocking trackers, intrusive ads, and protecting the user from mass tracking.

In Opera , the centerpiece is Aria, its own assistant. It runs in the sidebar, is also integrated into Opera GX, and offers image generation, quick commands for managing tabs, and direct access with keyboard shortcuts. Opera boasts that you don't need to log in to use Aria, although in practice the experience is enhanced if you link an account. To see how Opera is integrating other layers of AI into the browser, check out how it integrates Google's Gemini AI into its product.

Finally, there's the curious case of Google Chrome . Despite Google leading many areas of AI, its main browser has lagged far behind in generative integration. For now, the most visible feature is the ability to type @gemini in the address bar and send prompts directly to the model. Google has announced a future panel dedicated to Gemini and more features related to security and productivity, but for the moment, the integration is quite limited and hasn't yet been consistently rolled out on Android or iOS.

Intentionally “outdated” browsers: Vivaldi, Tor, LibreWolf and others

Faced with this frenzy to cram AI into every corner, some browsers have decided to stand firm and operate without generative models . Not because they can't integrate them, but due to privacy principles, rigor, or simply because they don't want to inflate the product with features many users don't need.

Vivaldi , developed in Europe and named after a composer, has stated on several occasions that it does not intend to integrate LLMs into its core functions "at least until the technology is more rigorous." Its reasons are clear: to avoid copyright issues, reduce the risk of hallucinations and biases, and maintain a browsing experience that does not depend on responses generated by opaque models.

Their strategy focuses on offering a highly customizable browser with a feature-rich interface, integrated VPN, tracking blocking, email client, RSS reader, and calendar . It positions itself as a haven for those who want power and control without AI intruding at every turn.

Tor Browser is in a league of its own. It's designed for those who prioritize privacy and anonymity above all else . It relies on the Tor network, encrypting traffic in layers (onion routing) to hide your IP address and prevent systematic tracking. It also includes ad blockers and tools to minimize fingerprinting.

In this context, it makes perfect sense that Tor doesn't want to hear anything about integrated AI : any system that collects data or maintains context clashes head-on with its philosophy. It's not a browser designed to automate your shopping or recommend restaurants, but to protect you in highly sensitive environments.

LibreWolf is a fork of Firefox that opts for a hardened configuration: no telemetry, advanced anti-fingerprinting protections , automatic session deletion, aggressive content blocker, and privacy-focused search engines like DuckDuckGo, Searx, or Startpage.

Its interface is quite reminiscent of Firefox and, to some extent, Edge, but under the hood, it's much stricter with data. It doesn't integrate AI or rely on browsers that already have it embedded by default. It's the ideal option for those who want a Firefox fine-tuned for extreme privacy without the AI ​​layers that Mozilla has been adding.

Avast Secure Browser, for its part, is part of the well-known antivirus ecosystem and is built on Chromium, but with a focus on proactive security. It offers Bank Mode, which isolates sensitive sessions to prevent keyloggers and spyware, integrated ad blocking, anti-tracking, and web shields that analyze potentially malicious sites or phishing attempts.

Here, a chatbot would be more of a hindrance than anything else: the priority is examining links, blocking threats, and maintaining a clean connection. For those looking for a robust browser without built-in AI , it can be an interesting alternative, although less well-known than others.

In addition to these, there are more niche options like Zen Browser, Falkon, and Maxthon , which also forgo virtual assistants and opt for a more traditional experience. They don't have the media presence of the big players, but they broaden the options for those who absolutely do not want AI in their browser.

A clear map is emerging between these two groups: on one side, browsers that want to be co-pilots or even drivers of your digital life (Atlas, Comet, Dia); on the other, solutions that prefer to be discreet windows to content, without intervening too much (Vivaldi, Tor, LibreWolf…). In the middle, the big names are trying to balance innovation with not alienating their user base.

The point of equilibrium will depend greatly on the individual: some will want their browser to plan trips, summarize reports, and fill out forms without blinking, while others will value staying behind the wheel, even if it means sacrificing some convenience. The only certainty is that the browser has ceased to be a mere frame for web pages and is gradually becoming the center of the battle for our attention, our data, and how we interact with information.