OpenAI launches ChatGPT Work agent for multi-hour projects
Reviewed by Helen Jones
On 9 July 2026, OpenAI announced ChatGPT Work, an agent inside ChatGPT powered by Codex and GPT-5.6. The company says it can take action across apps and files, stay with a project for hours, and turn a goal into finished work rather than a single chat reply. The post drew millions of views and sits in a wider wave. Claude Cowork expanded to mobile and web on 7 July, and industry coverage is grouping ChatGPT Work, Claude Cowork, and similar products as knowledge-work agents. Full access rules, costs, and desktop details remain on official product pages, not in the launch posts.
Until now, most professional AI use still meant short chats: paste context, get a draft, then spend time fixing tone, facts, and missing business detail. Multi-hour work stayed with the human, even when AI helped at the edges. What changed is the public framing. Major vendors are selling agents that claim to hold a goal across apps and files for hours. That raises the real risk for managers: teams may treat unfinished agent output as finished work, and one-off demos will not become reliable workflows unless someone defines context, review steps, and handoffs first.
Analysis
Treat this as a category signal to watch, not a reason to hand a live project to an agent yet. Pick one multi-hour task you already own, write a one-page workflow blueprint with required context, quality checks, privacy limits, and a human review step, then only test ChatGPT Work or Claude Cowork against that blueprint.
Source note
Pulse published by Collab365 Spaces, reviewed by Helen Jones on . Cite as "OpenAI launches ChatGPT Work agent for multi-hour projects", Collab365 Spaces. 2 sources referenced.