The people who made ChatGPT simple just lost to the people who made Codex complicated.
The Summary
- OpenAI reorganized in May, putting Codex leader Thibault Sottiaux in charge of ChatGPT, the API, and platform teams — merging consumer and enterprise products under one roof
- John Gruber's take: they promoted the wrong people. ChatGPT succeeded because it was focused. Codex is "sprawling and confusing."
- The move signals a bet on complexity over clarity just as AI products need to get simpler, not harder to use
The Signal
OpenAI's May restructuring handed the keys to Thibault Sottiaux, the exec who built Codex into what the company calls "one of the fastest-growing products of all time." Codex, the API, and ChatGPT now sit under one product umbrella. The official story: these products are converging anyway, so why not converge the teams.
The stated logic is that Codex is "increasingly powering" consumer and enterprise offerings, which are gaining autonomous task execution. They're building a "super app" that mashes ChatGPT, Codex, and the Atlas browser into one desktop application. More features, more surfaces, more integration points.
Gruber's read is sharper: this is the complexifiers taking over from the simplifiers. ChatGPT won because it did one thing well. You typed a question, you got an answer. No manual, no setup, no decision paralysis about which endpoint to hit or which model to invoke. Codex, by contrast, is a developer tool with a thousand knobs and a learning curve.
"Instead of putting the eggheads from Codex in charge of ChatGPT, they should have put the product-minded people from ChatGPT in charge of Codex."
The difference matters because we're at an inflection point for AI products. The tech works. The models are good enough. The bottleneck now is interface design and user experience. Can a normal person, not a developer, actually use this thing to get work done. ChatGPT answered yes. Codex answered "well, if you read the docs and understand REST APIs and token limits and..."
What happens when you put the Codex team in charge of ChatGPT:
- More options. More modes. More things to configure before you get an answer.
- Feature creep disguised as "unification" — the super app as a Frankenstein's monster of use cases
- Death by a thousand product decisions, each logical in isolation, collectively incoherent
This isn't hypothetical. We've seen this movie. Microsoft did it with Windows. Google did it with Google+. Apple almost did it in the 1990s before Jobs came back and killed 70% of the product line. The pattern: engineers who love complexity get promoted, and they build what they would want to use, not what regular humans need.
The Implication
If you're building an agent product right now, the lesson is brutal: simplicity is harder than complexity, and the people who can ship simple products are rarer than the people who can ship feature-rich ones. Promoting based on growth metrics or technical chops misses the point. The question isn't who built the fastest-growing product for developers. It's who built the product your mom could use.
Watch what happens to ChatGPT over the next six months. If the interface starts to fracture, if new features pile up without clear hierarchy, if you find yourself hunting through menus to do what used to take one click — that's Codex brain infecting the product. And if OpenAI can't hold the line on simplicity, that's a wedge for Anthropic, Mistral, or whoever else is willing to just ship a clean chat box that works.