Last night before bed, I saw a news flash while scrolling my phone — OpenAI is integrating ChatGPT, Codex, and their Atlas browser into a single desktop “super app.” The engineering and applications team is leading the integration.
My first reaction was: finally.
But thinking about it more carefully, the interesting part isn’t the product itself — it’s the strategic shift behind it.
From Doing Everything to Doing Just Two Things
Just a few days ago, The Wall Street Journal reported on an all-hands meeting at OpenAI. Applications lead Fidji Simo said something pretty direct: “We cannot miss this moment because we are distracted by side quests.”
In plain English: they can’t miss the main quest because they’re running around doing side missions.
Looking back at OpenAI in 2025, they really were doing everything — Sora video generation, Atlas browser, a hardware device with Jony Ive, e-commerce features, agent mode. And the result? According to reports, agent mode lost 75% of its users because nobody understood what it was supposed to do.
Now they’ve decided to cut the noise and focus on just two things: programming tools and enterprise clients.
Why “Programming”?
Because Anthropic came knocking at their door.
The numbers are pretty stark: in enterprise customer first-time AI purchases, Anthropic now beats OpenAI over 70% of the time (73% according to Ramp data). Claude Code has basically become the standard for enterprise developers, and Cowork is extending its reach to non-technical users on the desktop.
OpenAI’s ChatGPT still dominates the consumer side — paid users are 8x Claude’s, 4x Gemini’s. But enterprise is where the real money is, and the stickiness is high. Once a company’s CI/CD pipeline is hooked up to a certain AI Agent, the cost of switching is massive.
So OpenAI started accelerating. They launched the Codex desktop app (macOS) in February, expanded to Windows in early March, and on March 19th acquired Astral — the company behind Python toolchain uv and Ruff.
The Astral acquisition is a telling move. They don’t just want to do AI code writing — they want to own the developer toolchain. When you’re using their linter and package manager, where else are you gonna go?
The Logic of the Super App
Cramming ChatGPT, Codex, and a browser into the same desktop app — on the surface it’s “integration,” but what’s really happening?
They’re making an entry point.
The new Codex is already a true autonomous software engineering agent — capable of independently writing features, fixing bugs, running tests, and submitting PRs. It has over 2 million weekly active users, running on GPT-5.3-Codex and GPT-5.4 under the hood. Add Atlas browser, which lets agents automatically operate on the web (booking tickets, shopping, filling out forms), plus ChatGPT’s general conversation capabilities — together, what they’re building is an “AI operating system.”
This is pretty similar to the logic behind WeChat mini-programs. You open one app, you can do everything, no need to leave. The difference is, WeChat relies on social stickiness, while OpenAI relies on AI capability stickiness.
Sam Altman said something when Atlas launched last year: “This is a once-in-a-decade opportunity to rethink the browser.”
Looking back, he wasn’t talking about the browser — he was talking about the entire way humans interact with computers.
Everyone’s Racing for the Same Spot
It’s not just OpenAI making moves. On the same day (March 19th), Bloomberg reported that Google started testing a standalone Gemini macOS desktop app, with “Desktop Intelligence” features — capable of perceiving what’s on your screen.
Anthropic already has the Claude desktop app and Cowork, Perplexity has Comet browser, and even Apple is in talks with Google about using Gemini to rebuild Siri.
Every company is racing for the same spot: that “always-on AI entry point” on your computer.
This reminds me of 2007. Back then, everyone was fighting for the mobile app entry point. Now the fight is for the desktop AI entry point. The difference is, in the mobile era you might install a dozen apps, but this AI desktop assistant spot — there’s probably only one or two winners.
One More Thing
Internally, OpenAI is reportedly discussing a Q4 2026 IPO. Anthropic is also preparing their S-1 filing. Both are racing to go public first, because the first one out will set the valuation benchmark.
So this “super app” isn’t just a product decision — it’s part of the investor story. “We’re not a chatbot company anymore, we’re an AI platform company.” That sentence means completely different things for valuation.
But for me, the most interesting observation is this — OpenAI spent a year trying everything, and now they’re doing subtraction.
In an AI era where everyone feels like they need to add more, the biggest company is starting to subtract.
This shift itself might be worth more attention than any product launch.
$14.90 · 8 chapters + 6 templates
Learn More →