- AI Browsers are here
- what do they do?
- who are they for?
- where are they going?
- what are the risks?
- what does winning mean?
- Bear case: Social media browsers
- what they did
- how they failed
- what did they get wrong?
- social media was “different”: developers had earliest versions, the protosocial media of forums, mailing lists, and IRC. college kids got it in its pure, refined form in Facebook (2004) and MySpace (2003)
- Bull case: Chromium Renaissance
- Google did the impossible in the late 2000s and supplanted the sleeping giant Microsoft
- Perplexity, et al. are poised to disrupt the incumbent
- history rhymes: chrome was built on webkit; AI browsers are build on chromium
- Chrome made websites faster and more reliable, users benefited from switching even if they never used google services
- so is the bull case that web browsing will be enhanced by AI so much that users will want to switch? or professional users to pay for it?
- BUT
- Google has lowkey remade themselves into an AI company, despite product execution fumbles.
- Chrome was a technology wedge, a fundamentally better approach to building a scalable, high performance browser platform.
- Chrome differentiated by shipping faster than any browser vendor had ever considered.
- Will users care?
- What do users want? What is the strategy?
- the web is the ubiquitous lowest common denominator
- charging for a browser vs giving it away for free
- native apps (powered by web technologies, often including browsers directly) prove that users care about it being easy to use your app
- Do normal users spent much time thinking about their browser?
- How will these products be received?
- Like Cursor? PLG via forking and charging money
- arbitrage that people are willing to pay for something that they can get for free elsewhere?
- developers were the first to get the taste of AI
- Like RSS? redistribution with ads
- AI is a new way to re-interpret and re-contextualize existing content
- Like Google?
- Like Cursor? PLG via forking and charging money
- What happens in 1 year? 5 years? 10 years?
- Who pays the check?
- Google has bundled
1. AI browsers are here
Several AI browsers have been released this year, what’s all that about?
Since the dawn of the Internet, browsers have been prohibitively expensive to make and maintain. Despite this truism, over the past several months the next generation of browsers infused with AI have begun to ship to customers. They’re not yet mainstream.
The Browser Company shipped Dia and is rapidly iterating. OpenAI is rumored to be working on a browser. Perplexity has released Comet.
What do they do?
AI features in browsers are not a new invention in 2025. Brave browser launched their AI integration Leo in 2023. The model capabilities that have arrived in the last 12 months are new. Reliable tool calling, reasoning tokens, and long context windows make a meaningful difference when interacting with content from the web.
Some of the common features are
- ability to summarize the current page
- smart history, open-ended search over what you’ve browsed
- chat with tool calls
- prompt templates
- deep research
- automated browsing
- document creations
- multi-modal capabilities: voice interactions, image recognition content generation
- context management: projects/folders, memories, skills
In some ways, it simplifies down to better search. When you type something into the browser, the next thing that happens is more substantial than it was in the past. The are new capabilities available as almost hidden commands: if you know what it can do, you can go very far.
Users discovering features is largely the bottleneck.
lots of opportunity to improve browsing history, tab management, search/chat history the key will be doing this in a way that doesn’t add more concepts. selfishly, I want folders for tabs and that’s about it. I can do the rest. But I get that most users will never self organize? so pull back the Arc feature of making suggested folders for them. AI should be self-organizing.
does adding some new chrome around a web page change anything?
(sidebar: why is Arc not the vehicle for this? They’ve discussed strategy)
2. Bear case: Social media browsers
There was an earlier VC funded frenzy to reimagine the browser around a technology product with explosive growth: social media browsers.
The theory was simple: integrate sites Facebook, Blogger, MySpace, and Twitter into your web browser, so that you never had to stop what you were doing to update to your social media feeds. Unfortunately, the idea failed to capture a meaningful market share. Furthermore, the stability of the integrations were at the mercy of the openness of the social media platforms whose profit motive places high value on users visiting their apps, not someone else’s.
Do you remember Flock? Or Rockmelt? The last time VC tried to muscle in to the browser market on the back of Chromium. I remember hating them at the time, already preferring Chrome from its earliest days. Chrome made the web feel like it was
- reading
3. Bull case: Chromium Renaissance
In 2006 Google began work on a browser for the next wave of dynamic, AJAX-powered web apps like Gmail, Flickr, Facebook, and Google Maps. The Chrome browser released as a beta in 2008 followed by a 1.0 release in 2010. By the end of the decade it had swallowed 70% of the desktop browser market.
Google disrupted Microsoft and obliterated Mozilla.
The History of Google Chrome Browser: Primary Sources and Key Player Accounts - Bench It’s hard to not just repeat the research here verbatim when it has 60+ citations.
Could a similar displacement happen again? If a product offers a compelling enough experience, a better version of the internet, users will trust them to be their web browser.
4. What do users want? Strategy in the age of viral AI products
You generally don’t know what users want until you build it. One exception to what you can know in advance of user testing, is that users want things to be as simple as possible, but no simpler.
How do you present users with new capabilities that are in principle hard to describe, because they are so open ended?
AI is unruly because users don’t know what it can do. (https://x.com/emollick/status/1947089159644180794)
- reading
- Using AI Right Now
- Ask LukeW, “better to be vague and correct than specific and incorrect”
- LukeW | Prompt Building User Interfaces
Subsection of this, what do developers want? Will AI be a new layer to disintermediate between the sources of content and their presentation? Is is more like RSS than social media? Will the browser ever return to the original vision of being self-contained publishing platform?
Somewhere between 2024 and 2025 Cursor became impossible to ignore. Will the AI browsers capture the same moment?
(talked with Kevin a bit [[2025-07-22]]. his question was “what’s the strategy of everyone chasing down browsers right now? Like what’s Perplexity’s strategy? Get a critical mass of users to switch over and become too big to fail?” I thought about this some more last night. maybe AI browsers are an attempted to make the )
- mucho required reading
- find all posts from Browser Company, Perplexity, and OpenAI that mention their browser AI Browsers Research: Official Statements from The Browser Company, Perplexity, and OpenAI - Bench
Will users stick? What’s the cost of migrating browsers for a user? My guess is pretty low. What’s the cost of maintaining a browser, even as a fork of chromium? My guess is pretty high.
[[Technology Mirage]]
The maintenance race concept: whoever can embrace the long-term duty cycle of maintenance will live to fight on, anyone who can’t is doomed.
Things are happening, but you don’t know what they are. You’ll only know 10 years from now, when you can look back and put your finger on it, bisect back to the moment when something actually happened that has long term consequence. Mostly in the moment it’s nothing you can sense directly, like a tachyon whipping through the planet but not interacting on it’s way to who knows where, only that it’s point of origin is cosmic and far off.