OpenAI's July 9 livestream was billed as a model launch. The piece it leaned on hardest was not the model. It was ChatGPT Work, a new agent inside ChatGPT that is meant to do the task, not just answer a question about it.
What ChatGPT Work is
According to OpenAI, ChatGPT Work can take actions across your apps and files, stay with a single project for hours, and turn a goal into finished work. Instead of a reply you then have to act on, the pitch is that you hand over the whole job and it comes back done. Sam Altman called it the biggest of the day's announcements.
Built on GPT-5.6 and Codex
ChatGPT Work runs on the new GPT-5.6 model and Codex, OpenAI's engine for writing code and operating software. Folding those together means the same assistant can research, write, and actually click through applications on your behalf. OpenAI also rebuilt the ChatGPT desktop app with Codex inside it, and added a Sites beta that turns a chat into a hosted web page.
That is a different product than the chatbot most people know. It is ChatGPT trying to become something closer to a coworker.
The obvious comparison
The clearest way to read ChatGPT Work is as OpenAI's answer to Anthropic's Claude Cowork, which shipped the same idea a couple of weeks earlier: hand a whole task to an agent that works across your tools. The two companies are now racing to own the same shift, from an assistant you chat with to one you delegate to.
This is OpenAI's direct answer to Anthropic's Claude Cowork: the same idea of handing a whole task to an agent that works across your apps and files, now built into ChatGPT.
I have been saying for months that Codex is excellent at knowledge work, not just coding. Now that engine is built into ChatGPT through ChatGPT Work, sitting in the same place you already chat, search, and write, so the whole workflow lives in one spot.
If you want to go deeper, here is my full course on using Codex for office work, which has shifted a little now that Codex lives inside ChatGPT, though the same principles hold.