OpenAI stopped selling a chat box and started selling a worker
Yesterday OpenAI launched ChatGPT Work, a GPT-5.6 agent you hand an outcome to, not a prompt. It gathers what it needs across your connected apps, breaks the job into steps, and comes back with finished sheets, slides, docs and sites, sometimes after working for hours. It ships inside a rebuilt desktop app that puts Chat, Work and Codex on every plan, including Free, on Mac and Windows. It's rolling out now to Pro, Enterprise and Edu, with Plus and Business over the next few days.
What it actually means
The product form of AI just moved from 'a chat box you steer' to 'an agent you delegate to.' First, the free desktop app now hands your non-technical colleagues an agent that builds real deliverables, so the gap between people who can prompt and people who can ship just narrowed inside your own team. Second, this is a straight answer to Anthropic's Claude Cowork — which means 'an agent that finishes the job' is now table stakes, not the differentiator. The move this week isn't to watch the keynote. It's to hand it one real, end-to-end task you'd normally do yourself and watch where it breaks the wrong assumption it makes on hour two is the thing worth learning, not the polished demo.
The product form of AI moved from a chat box you steer to an agent you delegate to.
Source: 9to5Mac · Friday 10 July 2026