OpenAI is collapsing the line between conversation and execution, turning the chat box from a product into the substrate for the next default consumer AI surface.
OpenAI has folded its coding agent, Codex, into the ChatGPT interface. The product move is small. The category move is not.
For ten years, the chat box was the default surface of consumer AI: a text field, a model, an answer. It was the product. In the week of July 9, 2026, OpenAI completed a long-trailed reorg that puts Codex, the company's coding-focused agent, inside the same ChatGPT shell that has carried the brand since 2022. The same window saw the company ship GPT-5.5 in April and follow with GPT-5.6 in July. The cadence and the reorg landed together, and they read as one move: the chat box is no longer the destination.
The deeper argument belongs to Stratechery's Ben Thompson, who frames the merge as the moment OpenAI stops treating chat as the product and starts treating it as the substrate. The surface stays. What lives behind it shifts from "answer the prompt" to "execute the work." Thompson calls the larger pattern a "super app": one OpenAI-owned surface that absorbs browsing, documents, code, and, eventually, anything else an agent can do on the user's behalf. The architecture is already taking shape inside OpenAI's product, and the company is reportedly steering its IPO narrative around the same frame.
Secondary coverage has run in the same direction. Business Insider's July product tracker treats the Codex-into-ChatGPT move as a flagship release rather than a quiet integration. Fortune's architectural read traces the surface back to the same super-app logic. TechCrunch's April GPT-5.5 launch carried the framing in its headline. The wire is not contesting the read. It is also not naming the category underneath it.
The honest read is not that OpenAI is abandoning chat. Billions of users will keep typing questions into a box. What is changing is what the box optimizes for. A model that answers a prompt is a different product from an agent that opens a file, edits a spreadsheet, runs a shell command, and reports back. OpenAI has decided the second product is the higher-margin, higher-stickiness, higher-platform one. The Web Archive mirror of Thompson's piece preserves the chain of product decisions behind the call for readers who want the argument in full.
Two caveats belong in the read. The category-boundary framing is Thompson's argument, and it is partly a forecast about how the rest of the industry moves next: Google, Anthropic, and the open-weights camp are all building agents that act, and the consumer surface that hosts them is the next platform contest. And the chat surface served a category of users who were never going to write a function or run a tool. An OpenAI that optimizes for the agent surface may not optimize for them. The general-purpose assistant the world met in 2022 is being narrowed, even as the company says it is being broadened.
The next tell is product cadence. If OpenAI ships Codex-style execution into the default ChatGPT experience for free-tier users, the category shift is operational, not rhetorical. If it stays gated behind a paid tier, the super-app story is a pricing story wearing a product one. The wire will report the next release either way. The frame to keep is the one Thompson named first: chat was a category, not a destination.