OpenAI launches ChatGPT Work, an agent that stays with a project for hours to produce finished documents and web apps

OpenAI has introduced ChatGPT Work, a new agent inside its chatbot designed to execute complex, multi-step tasks across a user's applications and files. The release, announced on 9 July 2026, marks a significant push into workplace automation, moving ChatGPT beyond answering questions toward autonomously completing projects.

What ChatGPT Work can do

ChatGPT Work is built to gather information from a user's entire workflow, including desktop files, web resources and connected workplace tools. It can produce finished materials such as spreadsheets, presentations, documents and web applications. The agent is powered by the newly released GPT-5.6 model family (Sol, Tierra and Luna), which OpenAI says makes it state of the art at reasoning through multi-step tasks and following user templates and reference files.

Codex technology is the engine underneath. OpenAI notes that more than five million people use Codex every week, with over a million now using it for tasks outside software development. The company has merged ChatGPT and Codex under co-founder Greg Brockman, though Codex remains available as a separate view inside the redesigned ChatGPT app.

How it works across devices

Users can initiate a task from a smartphone and later track progress or continue working from the desktop app or web browser. The desktop version can access and modify local files and use a built-in browser to reach online tools. A new unified plugin directory connects ChatGPT Work to over 1,400 applications, including Slack, Microsoft Teams, Google Drive and SharePoint. Users can reference a specific app by typing "@" followed by its name in a prompt.

You remain in control of how ChatGPT works with you. You decide what it can access, when it should check in, and when it needs your approval before taking action.

Scheduled tasks and Sites

ChatGPT Work includes Scheduled Tasks, which can automate repetitive jobs on a timetable or when a monitored event occurs. OpenAI gives the example of converting new Microsoft Teams or Slack messages into updated documents or presentations and sharing the key changes with the team.

The agent also introduces a feature called Sites, which turns ideas, plans and data into interactive websites and web applications. These can serve as dashboards, project-tracking tools, launch calendars, prototypes or reports, and they stay updated as the underlying information changes.

Early customer use and enterprise controls

OpenAI points to early adopters to illustrate the range. Angela Ferrante, head of enterprise marketing at Zapier, used ChatGPT Work to build a repeatable system for reviewing thousands of leads per month, tracing customer touchpoints across the company's CRM and email to identify where follow-ups were needed.

ChatGPT Work helped us build a repeatable system for reviewing thousands of leads a month.

The company stresses that access to outside apps and files can be limited through its Compliance API or enterprise and admin controls. Users supervise the agent's progress, answer its questions, redirect work when necessary and approve important actions before they are executed.

App consolidation and Atlas sunset

With this release, the existing ChatGPT desktop app is being rebranded as ChatGPT Classic. The redesigned ChatGPT app now houses ChatGPT, ChatGPT Work and Codex in one place. OpenAI's dedicated Atlas web browser is being sunset less than nine months after its launch, with an updated ChatGPT Chrome extension taking over web-native tasks. The GPT-5.6 model family powering the agent had its release temporarily delayed by the Trump administration before clearing US government testing.