The Browser Wars

Today, everyone and their mother is building an agentic browser.

The Browser Co was bought by Atlassian

OpenAI launched Atlas,

Strawberry opened their beta,

Comet and Dia enabled agentic features,

BrowserUse launched Director.ai

Firecrawl announced they were building a browser.

After a decade of stagnation, browsers are suddenly interesting again.

AI has turned them from passive windows into active coworkers: interfaces that can see what we see and act on our behalf.

The rush to build the "agentic browser" isn't really about design or UX. It's about who will own the next layer of the internet's interface, and with it, the user.

Lessons from history

The First Browser War was the dot-com era battle where the innovator (Netscape) lost to the incumbent (Microsoft's Internet Explorer) through aggressive bundling.

Lesson: Distribution beats innovation. Bundling beats product superiority.

The Second Browser War began in 2004 when Firefox took on Internet Explorer with a faster, open-source, privacy-first browser and started gaining marketshare. Firefox monetized by directing traffic to search engines; at one point, google made up 85% of its revenue.

But Google, whose business depended on the browser, vertically integrated and won the market.

To avoid antitrust issues, it open-sourced Chromium the base on which Atlas, Dia, Comet, and Strawberry now run.

Lesson: Users don't pay for browsers. The player closest to the browser's core business wins.

The perfect storm

For a decade, browsers barely evolved: no monetization, high switching costs, and Chrome's dominance froze innovation.

That changed after 2022:

• The Browser Company built Arc, loved by the tech crowd for its design and power tools (easels hinted at generative ui).

• Google faces antitrust pressure over chrome's dominance.

• LLMs have started replacing google search as the web's front door.

Agentic browsers

Agentic AI took off at the end of 2024 with Anthropic's computer use demo and the launch of MCP. As models got better at multi-turn reasoning and tool use, browser infrastructure evolved too.

Browsers have effectively become Internet Computers: they sees what you are working on, making it the ideal substrate for agentic workflows.

Monetization

Inference is expensive and browsers haven't cracked monetization. That tension (high R&D costs, high variable costs and no clear business model) likely pushed The Browser Company to sell to Atlassian.

Yet if the early AI era taught us anything, it's that users will pay for intelligence (nobody paid for IDEs before Cursor).

The future of the internet

AI promises the personalized web. We won't browse websites, we'll get interfaces generated in real time, shaped by our preferences and intent (generative UI).

Owning the browser is one path to get there, but not the only one:

• Wabi is doing it through an app

• Nothing is doing it through a phone

• Mio is doing it through context infrastructure

Musings

If I were building a browser today, I would counter position: instead of building the browser for agents, I would build a humans only browser. With built-in proof of personhood, that browser would make the internet free (humans access content for free, bots pay a small fee) and extremely fast (30-40% of browser code is anti-botting).

Having extensively tested AI browsers, my default browser today is Strawberry (invite here). It's still rough around the edges but it's the best at doing an increasing amount of my busy work.