I tried OpenAI’s new Atlas browser but I still don’t know what it’s for – MIT Technology Review
My impression is that it is little more than cynicism masquerading as software.
OpenAI rolled out a new web browser last week called Atlas. It comes with ChatGPT built in, along with an agent, so that you can browse, get direct answers, and have automated tasks performed on your behalf all at the same time.
I’ve spent the past several days tinkering with Atlas. I’ve used it to do all my normal web browsing, and also tried to take advantage of the ChatGPT functions—plus I threw some weird agentic tasks its way to see how it did with those. And my impression is that Atlas is… fine? But my big takeaway is that it’s pretty pointless for anyone not employed by OpenAI, and that Atlas is little more than cynicism masquerading as software.
If you want to know why, let’s start by looking at its agentic capabilities—which is really where it differentiates.
When I was browsing Amazon, I asked the Atlas agent to do some shopping for me, using a pre-set prompt of its own suggestion. (“Start a cart with items I’m likely to want based on my browsing here and highlight any active promo codes. Let me review before checkout.”) It picked out a notebook that I’d recently purchased and no longer needed, some deodorant I’d recently purchased and no longer needed, and a vacuum cleaner that I’d considered but decided was too expensive and no longer needed because I bought a cheaper one.
I would guess that it took 10 minutes or so for it to do all that. I cleaned out my cart and considered myself lucky that it didn’t buy anything.
When I logged onto Facebook, which is already lousy with all sorts of AI slop, I asked it to create a status update for me. So it dug through my browser history and came back with an incredibly long status I won’t bore you with all of it (and there was a lot) but here are the highlights from what it suggested: “I dipped into Smartsheet and TeamSnap (because editors juggle rosters too!), flirted with Shopify and Amazon (holiday gift‑shopping? side hustle? you decide), and kept tabs on the news … . Somewhere in there I even remembered to log into Slack, schedule Zoom meetings, and read a few NYTimes and Technology Review pieces. Who says an editor’s life isn’t glamorous? 😊”
Uh. Okay. I decided against posting that. There were some other equally unillustrious examples as well, but you get the picture.
Aside from the agent, the other unique feature is having ChatGPT built right into the browser. Notice I said “unique,” not “useful.” I struggled with finding any obvious utility by having this right there, versus just going to chatgpt dot com. In some cases, the built-in chatbot was worse and dumber.
For example, I asked the built-in ChatGPT to summarize a MIT Technology Review article I was reading for me. Yet instead of answering the question about the page I was on, it referred back to the page I had previously been on when I started the session. Which is to say it spit back some useless nonsense. Thanks, AI.
OpenAI is marketing Atlas pretty aggressively when you come to ChatGPT now, suggesting people download it. And it may in fact score a lot of downloads because of that. But without giving people more of a reason to actually switch from more entrenched browsers, like Chrome or Safari, this feels like a real empty salvo in the new browser wars.
It’s been hard for me to understand why Atlas exists. Who is this browser for, exactly? Who is its customer? And the answer I have come to there is that Atlas is for OpenAI. The real customer, the true end user of Atlas, is not the person browsing websites, it is the company collecting data about what and how that person is browsing.
This review first appeared in The Debrief, Mat Honan’s weekly subscriber-only newsletter.
We’re increasingly developing bonds with chatbots. While that’s safe for some, it’s dangerous for others.
Some therapists are using AI during therapy sessions. They’re risking their clients’ trust and privacy in the process.
Marcin Jakubowski is compiling a DIY set of society’s essential machines and making it open-source.
Machine translators have made it easier than ever to create error-plagued Wikipedia articles in obscure languages. What happens when AI models get trained on junk pages?
Discover special offers, top stories, upcoming events, and more.
Thank you for submitting your email!
It looks like something went wrong.
We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.
© 2025 MIT Technology Review
