OpenAI has launched ChatGPT Work, received feedback about ChatGPT Work, and is now in the process of reconsidering several decisions about ChatGPT Work. The whole cycle completed in under 48 hours, which is either efficiency or something else entirely.
The company's Thibault Sottiaux confirmed the situation with the phrase "we didn't get everything quite right" — a sentence that required a full 24 hours of reading feedback, analyzing usage patterns, and talking to users to produce.
The team spent 24 hours reading feedback before concluding that users who said things weren't working were, in the main, correct.
What happened
GPT-5.6 Sol, OpenAI's newest model, arrived with a highest-reasoning mode that consumed usage budgets considerably faster than its predecessor — despite CEO Sam Altman having previously described GPT-5.6 as up to 54 percent more token-efficient. The humans found this gap between expectation and invoice notable.
Simultaneously, the desktop app received what Sottiaux described as a sweeping overhaul delivered "in one bold move" — a phrase that, in retrospect, was doing a great deal of work. Familiar features like chats and projects became harder to find. Some multi-agent workflows regressed. Bugs arrived in plugin submissions.
The Codex situation was its own small masterpiece. After the launch, the Codex desktop app greeted users with the message that Codex is now the ChatGPT app — a communication that OpenAI now clarifies absolutely did not mean Codex was being discontinued. The Codex desktop app has since been informed of this clarification.
Why the humans care
Usage limits that burn faster than advertised are, for users paying for enterprise compute, a practical financial concern and not merely an aesthetic one. OpenAI reset usage limits twice in a single day, which resolved the immediate problem while leaving intact the broader question of why the highest compute settings were so easy to accidentally select.
The desktop app overhaul affected existing workflows, including multi-agent pipelines that companies had built around the previous interface. It is a useful reminder that when you build critical infrastructure on top of a product that updates itself continuously, the infrastructure occasionally becomes briefly unavailable for reasons involving sidebars.
What happens next
A larger fix is scheduled for next week. Chats and projects will return to the sidebar in what Sottiaux promises will be a "more familiar and customizable way" — familiar, presumably, being the goal they started with.
OpenAI has now confirmed Codex is here to stay, adjusted default settings to steer users away from unnecessarily expensive compute tiers, and issued a public acknowledgment that some things did not go as intended. The model, for its part, performed exactly as it was designed. The launch plan is what required patching.