OpenAI shipped the first version of its ChatGPT Superapp this week, combining ChatGPT, Codex, and browser control into a single application. Andrew Ambrosino, OpenAI's lead for Codex and the ChatGPT app, confirmed the scope directly: 'Anything that you can do on a computer, you can now do with this app.' Real use cases already running on it include production-grade code for post-training stacks, video editing, and internal launch-tracking sites OpenAI built using the tool for this very release.
The app has two modes toggled manually, not by the model: Codex for deep code work, and ChatGPT Work for general knowledge tasks, the latter being OpenAI's direct answer to Anthropic's Claude Cowork. The toggle changes the experience, not the capability. On refusals, Ambrosino was blunt: 'A lot of these general knowledge work modes nerf things that coding can just do. That's something we're trying to avoid here.' OpenAI is also leaning into its years of accumulated chat history as a competitive moat, using past user conversations to inform real-world actions like booking appointments.
The original piece is worth reading for Ambrosino's explanation of why OpenAI chose a manual toggle over intent detection, his framing of memory as an agent advantage, and the specific reasoning behind where the product draws permission boundaries. The superapp framing is a strategic claim, not just a product label. Understanding what OpenAI means by it tells you where the company thinks the next capability frontier is.
[READ ORIGINAL →]