ChatGPT Work turns projects into long-running agent tasks

OpenAI introduced ChatGPT Work on July 9 as an agent for projects that extend beyond a single answer. The company said it can research, analyze information, use connected apps and files, and produce finished documents, spreadsheets, presentations, reports and Sites. The release reframed ChatGPT around completing a body of work rather than returning one response at a time.

The interface is designed to keep the user involved while the task runs. OpenAI said people can follow progress, answer questions, redirect the work and approve important actions. Those controls matter when an agent has access to business files or connected services, because the ability to reach a tool is different from permission to make a consequential change without review.

Scheduled Tasks add a second kind of delegation. According to OpenAI, a task can run once, repeat, trigger after an event or monitor for a change. The useful details are the trigger conditions, the data available at run time and the point at which a person must approve an external action. The launch materials describe the framework but do not prove that every workflow completes reliably without intervention.

OpenAI's release notes said the initial web and mobile rollout covered paid plans other than Free and Go. Pro, Pro Lite, Enterprise and Edu were listed first. Availability was rolling, so the presence of a plan name in the announcement should not be treated as a guarantee that every account, platform or region received the feature at the same moment.

The central product question is whether long-running work remains understandable and correct at handoff. Progress visibility, source quality, file accuracy, approval design and the ability to resume or redirect a job will matter more than the number of output formats. Those qualities require task-level evaluation with real project constraints.