What Is ChatGPT Work and How Can Operations Teams Use It?

ChatGPT Work is an AI agent that OpenAI added to ChatGPT on 9 July 2026. Instead of answering a question, it takes a goal and turns it into finished work: it collects information from the apps and files you connect, breaks the job into steps, can keep working on it for hours, and hands back a document, spreadsheet, deck, report or interactive site. For operations teams, its most useful job is turning scattered sources into a tidy report or decision file.

How is ChatGPT Work different from regular ChatGPT and Codex?

OpenAI's Help Center now splits ChatGPT into three experiences:

• Chat: quick questions and answers, search and brainstorming. The ChatGPT conversation you already know is still there.
• Work: an agent built for longer, multi-step jobs that need research, analysis and a finished output such as a document, spreadsheet, deck, report or site.
• Codex: software work such as writing code, debugging, running tests and working inside a repository.

That also answers the common "ChatGPT Work vs Codex" question. The two share the same underlying infrastructure and the same usage structure, but Codex is dedicated to software development. For the day-to-day work of marketing, finance, sales and operations teams, Work is the right entry point.

What can ChatGPT Work do?

Its capabilities fall into four groups.

It gathers information from your apps. Through connectors it can reach Slack, Microsoft Teams, Google Drive, SharePoint, email, calendar, CRM and project tracking tools. It decides which connector to use on its own, and you can point it at a specific source by typing "@" followed by the app name in your prompt.

It produces finished outputs. It can prepare documents, spreadsheets, decks and reports. With Sites, offered as a public beta, you can turn a dashboard, a project tracking page or an interactive report into a web page and share it with a link.

It runs recurring work on its own. With Scheduled Tasks you can run a job once, on a set schedule, or when an event happens. The event examples listed in the Help Center are a new Gmail message, a new message in a Slack channel and pull request activity on GitHub.

On desktop, it uses your computer. In the current ChatGPT desktop app, Work can operate on the local files and apps you allow and can pull information from websites through a built-in browser. Work on the web and mobile runs in the cloud and cannot reach the files on your computer directly.

Which operations tasks is it good for?

The task operations teams lose the most time on is turning information spread across different places into one consistent document. That is exactly where Work is strong. The scenarios below are practical examples based on the kinds of use OpenAI describes in its announcement:

• Weekly status report: read updates in Slack or Teams channels and tasks in the project tracker, then produce a status report in the same format every Monday.
• Month-end budget variance: compare the spreadsheet from accounting with the budget, explain the variances line by line, and draft a few slides for management.
• Project brief: read meeting notes, email threads and attachments and turn them into a brief with a clear goal, scope, owners and dates.
• Decision file: when you need to choose between two suppliers or two software products, pull the proposals, contract terms and information from their websites into one comparison table.
• Live dashboard: build a simple dashboard site from sales or order data that the team checks every day.

The same rule applies to all of them: Work only works from the input it has. It may miss an error in a spreadsheet or fill a gap with a guess. A person should check every file that goes to management.

How do you get started with ChatGPT Work?

OpenAI's advice is to give it a task you already know well. If you start with work whose result you can judge, you will quickly see where the tool is strong and where it is weak.

1. Open ChatGPT on the web, mobile or desktop and choose Work instead of Chat.
2. Create a project for related work; a project keeps the relevant conversations, files and instructions together.
3. Describe the result you expect, the files to use, the constraints and how you will check the output. For example: "Same format as last month's report, two pages at most, and the source of every figure named by its spreadsheet."
4. Follow its progress while it works, answer its questions and approve important steps.
5. If the result was useful, turn the same instructions into a Scheduled Task.

Keeping instructions and templates that work in a shared team document stops the tool from giving every person a different result.

Which plans include it, and how does usage work?

As of 2026, according to OpenAI's Help Center, Work is available on ChatGPT web and mobile on eligible paid plans. OpenAI announced that in the updated desktop app, Chat, Work and Codex are available on all plans, including Free. Which models you get and how much usage you have depends on your plan and workspace settings.

Work follows the same usage structure as Codex. Long, complex tasks use up a larger share of your plan allowance than a short chat. Plan details change often, so check OpenAI's own pricing page before you make a purchasing decision for your team.

What should you check on data security and privacy?

Giving Work access to more apps and files means more of your data is processed by AI. Clarify these points before you roll it out:

• Where it runs: Work on the web and mobile runs in the cloud. According to the Help Center, even for tasks that run locally on desktop, messages and task context may be stored in the cloud.
• Which apps it connects to: connectors work with the permissions they have in the connected app. Start with only the apps and folders the task really needs.
• Which actions require approval: Work can ask for your approval before important actions. Keep steps that affect the outside world, such as sending an email or changing a CRM record, behind an approval.
• Organisational controls: in business workspaces, admins can turn cloud and local use of Work on or off by role. Enterprise and Edu plans also offer the Compliance API and spending controls.

For files that contain customer or employee data, your obligations under the data protection law that applies to you remain in place; in Turkey that is the KVKK (Law No. 6698). Removing or anonymising personal data the task does not need, and recording which data is processed by which tool, keeps you on the safe side.

In short: a checklist for the first month

1. Pick one recurring task whose result you know well.
2. Connect Work only to the apps that task needs.
3. Write the instructions together with the output format and the checking criteria.
4. Put every step that affects the outside world behind an approval.
5. Check the first few outputs line by line, then turn the task into a scheduled one.
6. Share the instructions that work with the team.

If you want to connect AI agents to your own workflows and systems, take a look at our AI integration services. For processes that need a permanent record, such as approval flows, forms and reporting, our business management platform Nexfabra may be a better foundation. If you would like to talk about your own situation, get in touch.