OpenAI Just Launched an AI-Powered Browser

Yesterday, OpenAI launched Atlas, their AI-powered browser.

I knew it was coming for a while, but now it's finally here.

This is OpenAI's take on the AI-powered browser (similar, conceptually, to Perplexity's Comet). I think millions of us are going to move to an AI-powered browser -- because once you get used to it, it's impossible to go back. Browsers are such a fundamental part of our experience with technology that we often forget that they’re there. It’s our window into the world wide web.

So today, I want to break down:

• What Atlas actually does (and why the memory feature is a game-changer)

• The strategic implications (this is bigger than it looks)

• What this means for the future of ‘browsing’

—@dharmesh

What Atlas Actually Does

In simple terms, Atlas is a web browser with ChatGPT built in at its core. But it's not just "Chrome with a ChatGPT sidebar" -- it's fundamentally rebuilt from the ground up for an AI-first experience.

The feature that gets me the most excited is memory. Atlas can remember pages you've visited and create queryable context from your browsing history. This means you can ask questions like "Find all the job postings I was looking at last week and create a summary of industry trends so I can prepare for interviews".

This is huge. Your browsing history turns from a static list of URLs into living, searchable context that actually helps you get work done. This is a far cry from going through your Chrome history looking for that one article you know you read in the last few days — or was that last week? We’ve all been there.

ChatGPT also comes with you everywhere you browse. You can chat with whatever page you're on -- asking questions, getting summaries, or taking actions directly on the page, instead of the previous route of copy/pasting and sending screenshots back to ChatGPT to chat with pages.

I know a lot of people will be concerned about privacy with this level of access. The good news is browser memories are completely optional. You can toggle ChatGPT visibility on specific sites, view all memories in settings, archive the ones that aren't relevant anymore, or delete your entire browsing history to wipe associated memories.

Lastly, with Agent mode, ChatGPT can also take actions for you in the browser -- like adding recipe ingredients to Instacart, researching competitors across multiple sites, compiling team briefs from past documents, or booking appointments. So in that regard, we could called Atlas an “Agentic Browser” (which it is), but OpenAI is using “AI-powered browser”, and it’s their product, their party and only fair that it’s their positioning. AI-powered browser it is (for now).

This is currently in preview for Plus, Pro, and Business users. It's early, and it will make mistakes on complex workflows, but the direction is clear.

You can download Atlas here on macOS and sign in with your Free, Plus, or Pro ChatGPT account. Unfortunately, Windows, iOS, and Android aren't available yet, but they're coming soon.

Why This Is OpenAI's Most Strategic Move Yet

I’ve seen some early reactions from users across LinkedIn and X saying Atlas was underwhelming, especially since Perplexity’s Comet browser (launched back in July) has similar features.

In the broader scope, I think people (including myself) are having a hard time wrapping their heads around how big OpenAI really is. It's weird -- we've never seen a product grow this fast, ever. ChatGPT has 800M+ weekly active users.

But ChatGPT is a product you visit. You open it, ask a question, get an answer, then leave. A browser is where you live.

It's the primary surface area that billions of people use all day, every day. Email, work docs, research, shopping, entertainment -- everything happens in the browser.

By owning the browser, OpenAI doesn't just get visibility into what people do inside ChatGPT. They get visibility into everything a user does on the web.

There are two major moats I see with this launch:

1. Memory: Once Atlas builds up months or years of browsing context, switching to another browser means losing all that accumulated memory and intelligence. We've seen this same effect with ChatGPT versus other chatbots -- people stick with it partly because of their conversation history.

2. Data: With user permission, OpenAI can see how people actually use the web in real-time. That feedback loop makes their models smarter, which makes Atlas better, which attracts more users, which generates more data.

3. Partnerships: The one thing I’m hoping they will do, is open up this new ecosystem for third-parties. So we can build apps/agents that use this collective memory (on a controlled basis). The beginnings of something like this are already in place with the “Login with ChatGPT” idea. The dream is that one day the 1B+ users across ChatGPT and Atlas will be able to “OAuth” (connect) third-party applications that have some access to the collective memory about themselves. This would be the master stroke. Not sure it’ll happen, but a boy can dream.

This connects to the operating system play I wrote about after DevDay. ChatGPT became an OS for AI apps with the Apps SDK. Now Atlas is extending that vision further as an interface for the entire web.

Web Browsing is Changing Fast

I've been testing AI-powered browsers for some time now, and here's one thing I've learned: once you get used to having AI built into your browsing, it’s impossible to go back. Well…maybe impossible is not the right word. Unlikely is probably better.

It's sort of like switching from a smartphone back to a flip phone. Technically possible… but why would you?

The ability to ask questions like "What was that article I read last month about AI agents?" and actually get an answer changes how we interact with information fundamentally.

With OpenAI and Perplexity now going head-to-head, I don’t doubt others will launch similar products. I’m not sure who’ll win, but I do think OpenAI’s distribution moat is a significant advantage.

A couple things I’m watching:

1. Adoption rate among power users: If developers, knowledge workers, and creators start switching, mass market will probably follow.

2. How Google responds: Chrome still has 65% browser market share. They can't let this go unchallenged, but integrating Gemini deeply into Chrome while protecting their search ad business is... complicated.

My guess is that a year from now, most of us will be using some version of an AI-powered browser -- whether it's Atlas, Comet, or something Google scrambles to build.

The question isn't if this changes, but rather how fast.

—Dharmesh (@dharmesh)