OpenAI shipped the first version of its long-promised superapp this week, merging ChatGPT, Codex, and browser control into a single application. The product adds ChatGPT Work, a mode comparable to Anthropic's Claude, and gives the app direct control over a user's computer and browser. Andrew Ambrosino, OpenAI's lead for Codex and the ChatGPT app, confirmed the scope plainly: 'Anything that you can do on a computer, you can now do with this app.'
The app ships with a toggle between Codex, the code-intensive mode, and ChatGPT Work, the general-purpose mode. Ambrosino says the toggle does not change what the product can do, only the experience around it. OpenAI is also deliberately targeting over-refusal, a persistent failure mode in AI agents where the model declines to execute real-world tasks like booking appointments or making purchases. Memory from years of prior ChatGPT conversations feeds directly into the agent's decision-making, letting past user behavior inform live actions.
The original piece is worth reading in full for Ambrosino's explanation of why the product cannot yet infer user intent automatically, and for the specific production use cases already running on the app, including post-training stack code and internal launch tracking tools built by OpenAI's own team. The conversation surfaces real product constraints alongside the ambition, which makes it more useful than the launch announcement alone.
[READ ORIGINAL →]