OpenAI launches ChatGPT Work — a direct answer to Claude Cowork, and another sign chat is turning into a build tool
OpenAI released ChatGPT Work on 9 July, an agent that takes a goal and works autonomously for hours to hand back finished documents, spreadsheets and web apps, bundled into a new unified desktop app alongside Codex — landing just two days after Anthropic's Claude Cowork expansion, in the same 'AI teammate that builds things' category founders are increasingly searching for.
12 July 2026
OpenAI announced ChatGPT Work on 9 July, running on GPT-5.6. Rather than replying in a chat window, it takes a goal, pulls context from a user’s connected apps and files, works unattended for hours, and returns a finished artifact — a document, a spreadsheet, or a working web app — instead of a conversation. It ships inside a new unified ChatGPT desktop app that also houses Codex, OpenAI’s coding-focused agent, with early case studies naming Zapier, Virgin Atlantic and NVIDIA as users. Rollout started on desktop for all plans immediately, with web and mobile access following for Plus, Pro, Business, Enterprise and Edu tiers over the following days.
The timing matters more than the feature list. This lands two days after Anthropic expanded Claude Cowork to mobile and web — the same product category, from the two labs setting the pace in this space, in the same week. Both are chasing the same buyer: someone who wants an AI system to actually produce the deliverable, not just draft it, and increasingly that includes non-developers who want a working app or internal tool without commissioning a full build.
So what
For founders and product leads, this is worth reading as confirmation that “AI that builds the thing for me” is now a genuine two-horse product race, not a single vendor’s bet. That’s good news if you’re using these tools to accelerate early prototyping or internal tooling — competition here moves fast and prices tend to fall. It’s a different story once you need something that has to actually ship to customers, integrate with existing systems, handle real data, or stay maintainable past the first version: that’s where an autonomous agent handing back a finished-looking artifact and a production-grade build diverge, and where teams need to bring in people who know which gap is which before it becomes an expensive rebuild. If you’re weighing where AI tooling ends and a proper build needs to start, our AI-assisted development page covers how we draw that line, or get in touch to talk through your specific case.