The app that's supposed to be the future of AI-powered work just shipped as a monument to organizational chaos.
The Summary
- OpenAI released a new "super app" version of ChatGPT for Mac and Windows, but the upgrade path is so confusing that users with the old app installed get zero notification it exists
- Benedict Evans called it "a total mess" with unclear distinctions between projects, tasks, and chats, and UI decisions that make no organizational sense
- The old "Classic" app still ships for mobile and can coexist with the new app, revealing OpenAI's internal structure problem: the AI team is world-class, but the product org is a disaster
- The new app looks polished at first glance but is "playing dress-up as a big-boy app" without the coherence to match
The Signal
OpenAI just did something remarkable: they took the simplest, most focused AI interface in the world and turned it into a confused productivity suite that even tech industry veterans can't figure out. The new ChatGPT desktop app adds "Chat, Work, and Codex" modes, plus projects, tasks, templates, and plugin integrations. But as John Gruber discovered, if you have the old app installed, the built-in update mechanism won't tell you the new version exists. You have to know from outside sources that there's a completely different app to download.
Once you do download it, the confusion deepens. Why do chats get a floating window but tasks and projects don't? Why does selecting "plugins" show you "templates"? The Setup flow won't let you finish unless you connect Slack or Google Drive, creating a barrier that's more nagware than onboarding.
"It is sometimes observed that in companies dominated by internal politics, their shipped product reflect the company's org chart."
This is Conway's Law in action. OpenAI's research team keeps shipping frontier models. But the product team just shipped an interface that suggests they have no idea what ChatGPT is supposed to be:
- A chat interface for conversations with an AI?
- A project management tool competing with Notion?
- A coding environment rivaling Cursor?
- A workplace hub replacing Slack?
The answer, apparently, is "yes" to all of the above, implemented simultaneously with no coherent vision. According to sources adjacent to OpenAI, senior executives are consumed with FOMO about Anthropic, and only the AI research team is considered A-team. The product org is left to ship whatever features bubble up from internal politics rather than user needs.
Here's what's telling: the old focused ChatGPT app still ships for mobile, including iPad. That decision, treating iPad as "a big iPhone" rather than a real computer, reveals how little strategic thought went into platform decisions. Mobile users get the clean, simple interface. Desktop users get the kitchen sink.
The Implication
This matters because OpenAI is trying to build the default interface for AI agents, but they're proving they can't build coherent software for humans. The company that's supposed to be leading us into the agent economy just shipped an app that can't decide if it's Slack, Notion, or a chat window. When your core product becomes a reflection of internal dysfunction rather than external user needs, you're not building infrastructure for Web4. You're building the next Skype for Business.
If you're building on OpenAI's platform, watch how fast Anthropic and others ship coherent experiences. The API might be world-class, but the reference implementation just became a cautionary tale about what happens when product vision loses to org chart politics.