The Browser Redefined: How AI is Remaking the Web's Front Door
(and why Open source will play an important role)
When I first went online in the late nineties, my portal was a single icon: Netscape. It quickly became the most-used application on my computer, a window to a new world. Other browsers soon followed—Internet Explorer (1995), Opera (1996), Safari (2003), Firefox (2004), and the eventual titan, Google Chrome (2008).
But for 20 years, the browser itself saw surprisingly few real innovations. We got the omnibox (merging the search and URL bar), tabs, extensions, and behind-the-scenes improvements like sandboxing.
For years, the market has been sleepy and locked in place. The market shares have barely budged[1]:
But the emergence of powerful, generative AI is fundamentally changing this status quo. We are now seeing a wave of challengers leveraging AI to redefine what a browser is, including The Browser company (Arc and Dia), OpenAI’s Atlas, Perplexity’s Comet, Kosmik, or the privacy-focused Brave.
This shift has changed the browser from a simple viewer into a more complex, three-part stack: Engine + UI + AI.
The New Browser Stack
The Engine: Today, the market for browser engines is almost entirely an oligopoly: Google’s Chromium (powering Chrome, Edge, Opera, and Brave) Mozilla’s Gecko (powering Firefox), with Apple’s WebKit being the mandatory engine for Safari on its devices. All these engines are Open Source which makes it relatively easy for others to build differentiation on the other two elements of the stack. Another example where Open Source has won !
The UI (User Interface): While many UI innovations have been limited, AI is forcing a change. The new AI sidebar is becoming a standard feature for contextual assistance. Browsers like Arc are evolving the omnibox into a unified “Command Bar”—a single place to search, browse history, and execute browser commands like “Toggle Dark Mode” or “Create a New Space”. We’re also seeing quality-of-life features like AI-powered tab grouping and the wider adoption of Vertical Tabs (native in Edge and Brave) to better manage screen real estate on widescreen monitors.
The AI (The Agent) is the revolution. Integrating AI introduces contextual assistance, deep automation, and knowledge synthesis. This new capability can be broken into three “superpowers.”
1. Agentic Automation
The most transformative feature is the “Agent Mode,” seen in browsers like Atlas and Comet, which can act on your behalf, executing multi-step tasks that used to require manual clicks and navigation. For example “Find the best-rated flight to Paris next month under $800 and add it to my cart,” or “Read the top 5 reviews for this product and summarize the common complaints”. This extends to intelligent form-filling and managing your email or calendar by referencing data from other open tabs.
2. Deep Contextual Understanding
Integrating a Large Language Model (LLM) into the browser allows the AI to access and interpret your active web page. This enables the AI to instantly summarize long articles, explain complex terms, or help rewrite drafts. Its true power lies in capabilities like Perplexity’s Comet, which performs cross-tab synthesis—for example, comparing the battery life and price of items open across multiple tabs.
3. Integrated Writing & Creation Tools
The AI is now accessible in any text field, eliminating the constant copy-pasting between websites and a separate chat interface. Users can invoke the AI to polish, expand, or rewrite text directly where they are typing, whether in an email, a social media post, or a document.
Example of OpenAI Atlas browser: vertical tabs on the left and chatGPT sidebar on the right
Agentic browser will bring the Third Age of Online Shopping
These “agent-first” browsers herald what I see as the third age of online shopping.
The First Age saw the development of e-commerce. Incumbents slowly moved online while new digital-native players emerged.
The Second Age was about aggregation and advertising. As sites proliferated, price comparison engines and aggregators became necessary. This age was dominated by the online advertising market—a market now valued at over $730 billion[2]—which leveraged cookies to target customers, predict behavior, and power retargeting. I witnessed that firsthand when I was CFO of an Advertising company.
The Third Age is the AI Personal Shopper. The AI features in these new browsers mean each person will have a personal shopper that knows more about them and their shopping behaviour than ever before, dramatically changing the way we buy online. For any e-commerce business, gaining access to this AI-powered shopper is now the most critical goal. Open AI is integrating eCommerce in ChatGPT, adding instacart to its app platform: Instacart is trading its direct hold over the user experience for the massive reach of ChatGPT’s tens of millions of daily active users in the US. And last month, Amazon told Perplexity to get its agentic browser out of its online store… the battle has started!
These new browsers also create new monetization models. For years, browsers had two main revenue streams: leveraging user data (Google) and monetise them indirectly or being paid to set a default search engine (a key revenue source for Firefox and even Apple). Now, Brave or Opera are adding revenue streams through advertising (Brave’s opt-in model) and cashback features for user purchases, offering a direct financial incentive to users while still generating income.
With the new AI browsers, two models are likely to complement them:
Default AI Model: There is a possibility that the “default search engine” deal could be copied to a “default AI model” deal. Browsers with significant usage could potentially get payments from large AI providers like OpenAI, Google, or others to be the default brain, but this only makes sense if the AI company can replicate the search engine model where being the default provider brings additional revenues.
Affiliation and Attribution: The personal shopper could have “incentives” to direct a user to a certain website and receive affiliation money. This creates an opportunity[3] to generate revenue and a challenge to rethink attribution mechanism. OpenAI has recently announced a deal with Paypal to be able to process payments for example.
An Unsolved, Insecure Future?
This new paradigm raises tectonic questions for the entire web.
The websites we use today are optimized for human eyes, not AI agents, many websites make an extensive use of Javascript which an AI agent doesn’t process efficiently. If a large chunk of online activity comes from agents, website owners who optimize their code for an AI might attract more purchases[4]. Will we see two versions of every website: one for humans and one for AI? And will small e-commerce businesses be able to afford this complexity, or will we see an even higher concentration of power? Open standards like llms.txt already exist to help LLMs better “read” websites.
Then there are the AI models themselves. Each AI request from the browser generates inference cost which can quickly become sky-high, challenging the business models of these companies and pushing them toward smaller, more efficient models operating locally rather than in the cloud. A small model (3-billion-parameter vs the XX Billion parameters for ChatGPT) needs around 5GB of RAM, making it well-suited for consumer hardware and for many of the features offered by an AI browser (summarizing a web page or several websites e.g).
But if these models are embedded in the browser, they can be reverse-engineered, so it’s pointless to keep them closed-source . Just as browser engines became open-source, it’s likely embedded AI models will be more open too. We’re already seeing this with the rise of hyper-efficient small open-source models from companies like Meta and DeepSeek.
Open source also brings much-needed trust, which is paramount due to the significant risks of these new features.
Privacy: Are we comfortable having not just our browsing history but also purchases, thoughts, and inner questions held by one large company?
Security: These new systems can be vulnerable. In late October 2025, security researchers successfully demonstrated a prompt injection vulnerability in OpenAI’s new Atlas browser, a foundational, unsolved security risk where hidden instructions on a webpage can trick an AI agent into performing malicious actions.
Still many questions!
The deck is being reshuffled and it’s not clear what the new landscape will look like once the dust settles. Google seems to have a strong hand (browser, data, AI models, content, AI chips). OpenAI has the largest AI user-base and brand recognition. Amazon has quite a few pieces of the required puzzle but if AI Personal shoppers can access amazon’s inventory directly, it threatens their advertising and product placement revenue streams[5].
We saw that Open Source is present in many places (Browser engine, models, websites tools…) These changes pose security and privacy risks, Will there be more space for privacy-focused, transparent Open Source players? what space is there for smaller companies to innovate?
Google just introduced Disco, a browser capable of building simple applications from your open tabs to help you filter, merge, and repurpose data across various sites
25 years ago, the web browser emerged as the portal to content, services and revenue. But it was just that… a portal, important, but commoditised. Today, AI makes this portal much smarter and there much more important to control. By making a browser choice, each one of us is participating in redefining what it means to interact with the web.
OH
I am grateful to my partner at >commit Max Corbani, and to our advisors and friends, Frederic Jacobs and Frank Zayan. Their insights and conversations on this subject were invaluable.
[1] Source : Statcounter : https://gs.statcounter.com/browser-market-share
[2] Source: http://www.einpresswire.com/article/815407574/global-digital-advertising-market-outlook-2025-2034-growth-drivers-share-and-trends
[3] In particular : https://rakuteninternational.com/news/new-report-affiliate-is-gen-zs-preferred-monetization-model
[4] We can also expect to have a share of the web traffic done by Cloud based Browser that operate without human interaction, further pushing the need for website to optimize their content for AI agents.
[5] And without a browser, the AI personal shopper called Rufus is unlikely to take off