OpenAI launched ChatGPT Work on July 9 as a GPT-5.6-powered agent for longer work across apps, files, web pages, documents, spreadsheets, presentations, Sites, and Codex.
The product matters because the GPT-5.6 launch is no longer only a model and API update. ChatGPT is becoming the place where OpenAI expects users to assign work, connect tools, review progress, approve actions, and receive finished artifacts.
OpenAI says ChatGPT Work can gather information across apps and workflows, break larger projects into smaller steps, and stay with complex tasks for hours. It can create sheets, slides, docs, reports, Sites, and web apps, while carrying context through a project.
The updated desktop app is the sharper change. OpenAI says ChatGPT can use local files and apps, work through a built-in browser, and use computer control to click, type, and move files where needed. The Codex app is also merging into the new ChatGPT desktop app, while Codex remains available as a coding agent inside the same broader surface.
Work moves from prompt to project
ChatGPT Work changes the unit of use.
A normal chat request is a turn. A Work task is closer to a project: connect sources, collect information, create a first artifact, refine it, ask for approvals, and keep updating it when inputs change.
That is why Scheduled Tasks are important. OpenAI says Work can perform actions once, repeat them on a schedule, trigger on changes, or monitor for updates. Examples include refreshing a recurring agenda from Slack updates, checking dashboards each morning, or turning customer feedback into prioritized product ideas.
This makes ChatGPT Work closer to an operating layer than a text box. If it can read team context, use connected plugins, browse, operate desktop apps, and produce files, the main question becomes what it is allowed to do without a human in the loop.
The governance surface expands with the feature surface
OpenAI says Enterprise and Edu admins can manage who has access, what company context ChatGPT can use, which tools it can connect to, and what actions it can take.
That is not a footnote. The product invites users to connect email, calendars, CRMs, project trackers, file systems, browsers, and local desktop apps. Each added tool increases the value of the agent and the blast radius of a bad instruction.
OpenAI also points to the Compliance API for visibility into Work conversations and actions, plus controls for plugin access, browser use, network access, sensitive actions, and desktop agent policies. Auto-review uses OpenAI’s advanced models to review important connected-tool actions before they happen.
For companies, this turns ChatGPT Work into a rollout decision. Teams should not only ask whether the agent creates better decks or spreadsheets. They should decide which tools it can read, which tools it can write to, when approval is required, how scheduled work is logged, and how usage gets budgeted.
