OpenAI's help documentation now reads like a product manager's nightmare — three apps, two modes, one brand, zero coherence.
The Summary
- OpenAI consolidated ChatGPT into a "super app" that now includes Work mode, Codex developer tools, and the original chatbot — but conversations don't sync between desktop and web, and features vary wildly by platform
- The new desktop app ballooned from 159 MB to 1.5 GB by absorbing Codex, Work, and remnants of the discontinued Atlas browser into an Electron bundle
- Users can now run ChatGPT Classic, ChatGPT, and Codex simultaneously, creating a confusing product portfolio that suggests internal dysfunction more than strategic vision
- Meanwhile, OpenAI is hiring a product manager for family experiences, signaling a push into households even as the core product fragments
The Signal
OpenAI's own help center describes a product architecture that would make any PM wince. Work mode runs in the cloud on web and mobile. On desktop, it can access local files. But cloud Work conversations don't appear in desktop Work. Desktop Work threads stay local. Codex only works in the desktop app, not web or mobile, except you can trigger some Codex tasks from the mobile Remote tab, but those don't become chat history. Got it?
This isn't feature richness. This is three products wearing a trench coat.
"These three paragraphs sound more like a critic's scathing review than a guide to how to use it."
The rebrand makes the confusion worse. According to 9to5Mac's breakdown, the path forward looks like this:
- The old ChatGPT Mac app is now "ChatGPT Classic" (159 MB, native)
- Codex became the new ChatGPT desktop app (1.5 GB, Electron)
- The new ChatGPT includes both Work and Codex modes
- Codex users can keep the Codex icon, but the app will say ChatGPT
- You can install all three versions simultaneously if you hate yourself
This product strategy betrays the tension every AI company faces right now: build narrow tools developers love, or build ambient assistants consumers tolerate. OpenAI is trying both. In the same app. With different backends. That don't talk to each other.
The timing is telling. While engineering scrambles to merge incompatible products, OpenAI posted a job for a PM focused on families, caregivers, and older adults. That's the consumer play. But the actual product shipping is a developer-focused Frankenstein that absorbed a web browser (Atlas, now dead) and gained 1.3 GB in the process.
Compare this to the agent platforms coming out of Anthropic and the focused coding tools from Cursor or Replit. They picked a lane. OpenAI is trying to be Slack, VS Code, and Alexa in one Electron wrapper.
The Implication
For developers building on OpenAI's APIs, this should worry you. Product chaos at the interface layer suggests chaos in prioritization. If they can't decide whether ChatGPT is a consumer assistant or a developer IDE, what happens when those use cases need conflicting model behaviors or context windows?
For users, keep ChatGPT Classic installed. The native 159 MB app will likely work better for basic chat than the 1.5 GB everything-app. The super app is OpenAI trying to own your entire workflow before they've figured out which workflow that is. Let them cook, but don't eat yet.