I’m posting this because I think the Atlas → ChatGPT Desktop transition exposes a broader product issue that is worth OpenAI thinking about.
I’m not a disgruntled user looking for a reason to leave ChatGPT. Quite the opposite.
I’ve increasingly bought into ChatGPT as my primary environment for information processing, and I now spend a significant amount with OpenAI relative to my overall AI spend. Despite plenty of people telling me “go Claude, go Claude”, ChatGPT remains my preferred system.
Atlas was part of why.
OpenAI persuaded me to do something quite significant: change my behaviour and move my everyday browsing into the ChatGPT ecosystem.
I made Atlas my browser. I accumulated bookmarks there. External links opened there. When I was reading something interesting, ChatGPT and I were effectively looking at the same thing. The boundary between “browsing the internet” and “working with ChatGPT” was starting to disappear.
So when OpenAI announced that Atlas was being incorporated into ChatGPT Desktop, I thought: great. That is exactly where it should go.
Except the migration seems to have focused heavily on bringing across the browser technology without fully bringing across the browser experience.
A few examples I’ve encountered already:
-
Atlas could be my macOS default browser. ChatGPT Desktop currently can’t, so clicking an external link kicks me back out to Safari/Chrome.
-
Atlas made the page I was browsing feel naturally part of my interaction with ChatGPT. ChatGPT Desktop can apparently use the current page, but I now need to explicitly invoke that interaction rather than it simply being part of the shared browsing context.
-
I built up bookmarks in Atlas, but I haven’t found a documented way of simply migrating those into the ChatGPT Desktop browser.
-
More generally, the new browser feels like an impressive browser capability inside ChatGPT, rather than yet being a complete replacement for the browser OpenAI persuaded me to adopt.
None of these things sounds particularly significant on a feature list.
That’s precisely the problem.
User experience is where the rubber hits the road.
An extra click in something somebody does once a month is irrelevant. An extra click in something they do fifty times a day changes the product.
The best technology eventually disappears into the user’s behaviour. You stop thinking about the tool and simply do the thing you came to do.
Atlas had started doing that for me.
I could:
click → browse → see something interesting → ask ChatGPT about what we’re looking at → work with it → keep browsing.
I didn’t really think about where the browser ended and ChatGPT began.
That’s an extraordinarily valuable thing for OpenAI to have achieved.
The frustrating part of this migration is that the underlying technology may actually be getting better while the experience, for someone who genuinely adopted Atlas as their browser, has temporarily become worse.
There’s a commercial point here as well.
I think of technology ecosystems a little like battery-tool platforms. Once I’ve invested in the battery system, individual drills, saws and other ‘skins’ aren’t completely independent purchasing decisions anymore. The battery, charger and interoperability become part of the value of every additional tool.
Increasingly, ChatGPT is the ‘battery’ for me.
Conversations, memory, research, Work, files, browser context and agentic capabilities aren’t really separate products in my mind. They’re becoming one information-processing environment.
That ecosystem effect is incredibly powerful when everything works together.
But it also means continuity matters.
If you’ve persuaded a customer to change their behaviour because your products integrate so well, a migration shouldn’t merely ask:
“Did we bring the old product’s technical capabilities across?”
It should also ask:
“Can the customer continue living the way our old product taught them to live?”
In my case, the answer currently seems to be: not quite.
And that’s why I’d encourage the Atlas/ChatGPT product team to look at things like default HTTP(S) handling, bookmark migration and seamless shared-page context not as minor browser features, but as migration continuity.
The strategic direction makes complete sense to me. I would much rather have Atlas become part of ChatGPT than maintain two separate OpenAI products.
But please finish the last mile.
You successfully changed my behaviour. Don’t make me change it back.