Tech

OpenAI launches ChatGPT Work, an agent for long-running office tasks

ChatGPT Work, launched on 9 July alongside GPT-5.6, connects to workplace apps and works through multi-step projects to produce documents, spreadsheets, presentations and web apps. Paid plans get it first.

Abstract illustration of a wireframe globe above a glowing perspective grid, in deep blue and cyan.
Abstract illustration of a wireframe globe above a glowing perspective grid, in deep blue and cyan.

OpenAI launched ChatGPT Work on 9 July, an agent inside ChatGPT built to take on multi-step workplace projects rather than answer one-off questions. According to launch coverage from Bloomberg, TechCrunch and others, it gathers information from connected business apps, can keep working on a task for hours, and returns finished documents, spreadsheets, presentations and web apps. It runs on GPT-5.6, the model family OpenAI released the same day.

What ChatGPT Work does

The basic idea is delegation. A user describes an outcome, and the agent breaks it into smaller steps and works through them, VentureBeat and SmartCompany reported. Bloomberg described it as designed to handle complex tasks “for hours at a time”. The output is the ordinary material of office work: documents, spreadsheets, slide decks and reports, as well as websites and web apps.

To do that, ChatGPT Work draws on tools that users connect through plugins. Integrations named in launch coverage include Slack, Microsoft Teams, Gmail, Google Drive and calendar apps. The work happens on OpenAI’s servers rather than the user’s device: an OpenAI product manager described the setup to VentureBeat as “a virtual machine in the cloud that’s always on for you”.

Tasks do not have to be one-off. SmartCompany, IT Brief and eWeek all reported a Scheduled Tasks feature for recurring jobs, with examples such as summarising Slack conversations, updating presentations and monitoring customer feedback.

Staying in the loop

OpenAI is presenting the agent as something users supervise rather than set loose. According to SmartCompany and IT Brief, users can follow the agent’s progress as it works and approve key actions.

That matters because the agent’s usefulness depends on access. To turn a week of Slack threads and emails into a report, it needs to read those Slack threads and emails, and to act on a calendar it needs permission to change that calendar.

Who gets it, and where

Coverage was consistent on the rollout order. On web and mobile, ChatGPT Work is available first to Pro, Enterprise and Edu subscribers, with Plus and Business users following over the subsequent days. It also features in a revamped ChatGPT desktop app for Mac and Windows; IT Brief and eWeek reported that the desktop app’s Chat, Work and Codex modes are available on all plans, including Free.

Detail
Launched 9 July 2026
Model GPT-5.6 (offered in Sol, Terra and Luna variants)
Platforms Web, mobile, and the ChatGPT desktop app for Mac and Windows
First access on web and mobile Pro, Enterprise and Edu
Following days Plus and Business
Typical outputs Documents, spreadsheets, presentations, reports, websites and web apps

A wider reshuffle

The launch came with other changes to OpenAI’s apps. TechCrunch and IT Brief reported that OpenAI is winding down Atlas, the AI-powered web browser it launched in October 2025. According to TechCrunch, some of Atlas’s agentic browsing features are moving into the ChatGPT desktop app, which gains a more capable built-in browser, and into a new ChatGPT extension for Chrome.

ChatGPT Work also arrives in a crowded week for workplace agents. Bloomberg noted that OpenAI and Anthropic are competing to build more capable agents for a broader range of work, and two days before OpenAI’s launch, Anthropic expanded its Claude Cowork agent from the desktop to the web and mobile apps, starting with its Max plan.

What it means

Analysis. ChatGPT Work is OpenAI’s clearest attempt to take the agent pattern that has become familiar in coding tools, where software works through a task step by step and reports back, and apply it to general office work. The pitch is less about a cleverer chatbot and more about handing off a piece of work, then checking back later.

That shifts attention to questions a launch cannot answer on its own: how reliably the agent finishes long tasks, how clearly it shows what it has done along the way, and how comfortable organisations are granting an always-on cloud agent access to email, files and business systems. Starting with Pro and institutional plans before the much larger Plus audience gives OpenAI a period to see how the agent is used before it reaches most paying subscribers.

It is also notable that OpenAI and Anthropic made similar moves in the same week: both are now positioning cloud-hosted agents that keep working without the user’s device as a central part of their apps, not an add-on.

Product details in this article are drawn from the launch coverage listed below; each detail was confirmed in at least two of those reports.

Sources

See All

More on OpenAI

See All

Type to search. ↑ ↓ to navigate, ↵ to open.