ChatGPT Work Hits 10M Users, Set to Merge Into Chat by Year-End
TL;DR
- OpenAI released ChatGPT Work on July 9 and, together with Codex, has reportedly crossed 10 million users three weeks in.
- Greg Brockman has confirmed that Chat and Work will merge by the end of the year, on a ChatGPT base estimated to have crossed 1B monthly actives in June.
- Each Work task runs in its own cloud computer with a /workspace/scratch directory; Pro accounts get 8 CPUs, 20GB of RAM and a 64GB disk.
OpenAI shipped ChatGPT Work on July 9, and three weeks in, Shlok Khemani's breakdown at Latent Space puts a specific number on the traction. Work, together with Codex, has reportedly crossed 10 million users, on a ChatGPT base estimated to have crossed a billion monthly actives in June.
Work is, in Khemani's phrasing, an agent for knowledge work that you connect to the places you already work: Slack, email, Drive, calendars, CRMs, project trackers. Under the hood each task runs in a cloud computer of its own, with Pro accounts getting 8 CPUs, 20GB of RAM and a 64GB disk, and each thread getting a working directory under /workspace/scratch. Around the launch the old App Directory has been rebuilt as a Plugin Directory, and Work currently sits beside Chat as a separate mode inside ChatGPT.
The reason to pay attention beyond the launch numbers is what Greg Brockman has confirmed: Chat and Work will merge by the end of the year. Whatever primitives OpenAI is settling on inside Work — scheduled automations, proactive suggestions, remote browser control, plugin-driven skills — get inherited by a product with roughly a billion users. That is a short window for anyone building on top of ChatGPT to figure out where their prompts, integrations and workflow assumptions still hold.
The honest caveat is that this is one writer's read, and Khemani flags plenty of rough edges of his own: plugin discovery is weak, the cloud browser gets rejected as a suspicious session on some sites and times out on others where a local browser is fine, and useful features like scheduled heartbeats are not yet available everywhere. The 10M figure is reported rather than audited, and it bundles Work with Codex rather than isolating Work on its own.
What is worth watching is the shape of what OpenAI is describing. Less a chatbot with tools bolted on, more a persistent worker with its own filesystem, its own browser session, and its own sense of when to act without being asked. If Brockman's timeline holds, that becomes the default ChatGPT experience for a billion-user product in a matter of months.
Shared on Bluesky by 2 AI experts
Originally reported by latent.space
Read the original article →Original headline: Unpacking ChatGPT Work: the Agent for a Billion Users