OpenAI ships ChatGPT Work, an agent for multi-hour enterprise projects

ChatGPT Work is OpenAI's clearest bid yet to move from conversation to sustained execution. Rather than answering a prompt and stopping, the agent 'takes action across your apps and files' and can persist on a project for hours, translating a goal into completed multi-step work. It arrives powered by the new GPT-5.6 family and folds the Codex coding app into a unified ChatGPT desktop client, consolidating OpenAI's developer and knowledge-work surfaces.

The pitch is squarely enterprise: automate complex, multi-step office tasks end to end. That lands OpenAI in direct competition with Anthropic's Claude Cowork and AWS's agent tooling, and reflects the week's dominant theme — the industry racing to turn chatbots into autonomous coworkers.

Developers were skeptical about how much of this is real autonomy versus demo-ware: ChatGPT Work drew a 327-point Hacker News thread debating agentic productivity against hype. Wharton's Ethan Mollick offered a pointed caution to all labs branching from coding into general knowledge work: 'non-coders are not just dumber coders' — stripping options from a coding app doesn't make it better for knowledge work; users need more control and visibility, not less. The bar for 'stays on a project for hours' is reliability across app boundaries, which is exactly where agents have historically broken.