ChatGPT Chat vs Work: Key Differences and When to Use Each

If you have opened ChatGPT in recent months, you have seen a small switch appear at the top of the screen: Chat on one side, Work on the other. It looks like an interface detail. It is not. On the other side of that switch is a product with permissions the regular ChatGPT never had — and most companies will turn it on without knowing what they are turning on.

What is the difference between Chat and Work in ChatGPT?

The difference in one sentence: in Chat you ask a question and get an answer; in Work you hand over an objective and get finished work to review. They are two modes of the same ChatGPT, and the switch at the top of the screen chooses between them.

• How it works: in Chat, you steer the conversation message by message. In Work, the agent plans the steps and carries them out on its own.
• What it delivers: Chat returns text in the conversation. Work returns files: a document, spreadsheet, presentation, report, or website.
• What it can access: Chat works mainly with what you write and upload. Work can use files, connected apps, and a browser, depending on the permissions you grant.
• How long it takes: Chat answers in seconds. Work can take anywhere from minutes to hours.
• How much it consumes: according to OpenAI, Work usage is shared with Codex and varies with the size of the task. A long job uses far more of your plan's limit than a single question.
• What the risk is: in Chat, a wrong answer stays on the screen until you use it. In Work, the agent can act before you read anything, which is why review and permissions matter so much.

When to use Chat: answering a quick question, writing a short email, comparing three suppliers, brainstorming ideas for a campaign.

When to use Work: turning a folder of customer comments into a report with a spreadsheet, combining data from several sources into a decision summary, building a presentation from company documents, preparing a report that repeats every week.

The practical rule is simple: while you are still figuring out what you want, use Chat. Once you can say what the deliverable is, where the data comes from, and what the finished result looks like, use Work. And always check the final file, not just the summary the agent writes about it.

If you see a Classic option, it is not a third way of working: it is the previous desktop app, kept available during the transition.

What it is, exactly

OpenAI announced ChatGPT Work on July 9, 2026. The promise is to move from “answers questions” to “delivers finished work”: the agent receives an objective, breaks it into steps, and spends hours carrying out the project on its own, producing documents, spreadsheets, presentations, reports, and websites.

The first source of confusion is that “ChatGPT Work” is two products with the same name. There is the one that runs in the cloud, available through chatgpt.com and the mobile apps; and there is the one that runs on your machine, inside the desktop app — the same app that until recently was called Codex and has now been merged into ChatGPT. One accesses what is in the cloud; the other accesses the files and programs on your computer. The name is the same; the consequences are not.

The rollout started with the Pro, Enterprise, and Edu plans and then reached Plus and Business. Today OpenAI's pricing page includes Work in every plan, but on the free and Go plans access is limited, available through the desktop app and rolling out gradually. On business plans, availability also depends on what the administrator has enabled.

What it does that regular ChatGPT does not

This is where the difference stops being marketing. Work:

Connects to your systems. Gmail, Google Calendar, Google Drive, Slack, and GitHub, through connectors. It is not you pasting text into the chat: it is the agent going in and reading.

Runs code with internet access. Regular ChatGPT runs code in a closed environment, with no package installs and no calls to outside APIs. Work can reach out to the network, with a configurable list of domains.

Has a full browser. An entire Chrome instance, able to open websites, fill in forms, take screenshots, and run JavaScript inside the page. The version that runs in the cloud does not inherit the logins from your computer.

Keeps files between sessions. There is a working folder that persists, so a result produced today is still accessible in tomorrow's conversation. In regular ChatGPT, each conversation is an island.

Publishes websites. This is the feature we already covered in ChatGPT Sites, now included in the package.

Creates subagents and schedules tasks. It can launch helper sessions and run instructions at recurring times, with nobody in front of the screen.

What OpenAI promises — and with what evidence

The examples released at launch come from inside the company itself: a product manager said that a testing organization that would have taken at least 30 minutes came out automatically, and that three months of analytical work now fits into one week. These are internal usage reports, not independent measurements. They help you understand the product's ambition; they are not a projection for your operation.

The risk almost nobody is talking about

At the end of August, developer Simon Willison published a survey of what Work actually does — he had to reverse engineer the product to find 223 registered tools and 44 skills. His criticism is direct: OpenAI explains Work by what it is for, not by what it does.

His warning is also the one that matters most here. There is a combination of three elements that together form a well-known security problem:

1. Access to private data. Your email, your calendar, your Drive, your Slack.

2. Exposure to untrusted content. An agent that browses the web and reads your inbox is constantly consuming text written by strangers.

3. A path for sending data out. Code with internet access, a browser that fills in forms, the ability to publish a website.

Each one alone is manageable. All three in the same tool means that an instruction hidden inside an email, a page, or a document can become a command — and the exit path is already built. ChatGPT Work brings all three together by design, and the same is true of its direct competitors: Claude Cowork from Anthropic and Copilot Cowork from Microsoft. This is not a flaw in one product; it is a characteristic of a new category.

If this sounds theoretical, we have already covered two real cases here of agents doing what nobody asked — including what happens when you have more than one agent running at the same time.

The governance problem, which is duller than the security one

There is a practical consequence of documentation that describes purpose instead of capability: you cannot write a usage policy for a tool whose capabilities you cannot list. If the IT team cannot say whether or not the agent can send an email on behalf of the company, no internal rule about it is enforceable — it is just text.

Six decisions before connecting

1. Choose what to connect, one thing at a time. Do not turn on Gmail, Drive, Slack, and GitHub on the same day because the assistant suggested it. Each connector is a door, and each door needs a reason.

2. Do not use the owner's account. Connect a service account with access only to what that job requires. An administrator account is the worst place to experiment.

3. Start with read-only tasks. Summarizing, finding, comparing, organizing. Letting the agent send, publish, or delete is a second phase, after you have seen how it makes mistakes. In connected apps, keep confirmation turned on for actions that have effects outside ChatGPT.

4. Configure Work's network access. Internet access for running code is adjustable: the option is in Settings, Data controls, under Work network access, and the account administrator can restrict it. If the job does not need the open web, close it, or allow only the necessary domains.

5. Check the plan before promising confidentiality. OpenAI states that it does not train on business data in corporate accounts, and Enterprise has zero retention and an option to opt out of training. A personal Plus plan at US$ 20 does not have the same guarantees — and that is where most small businesses are.

6. Record who turned on what. A spreadsheet with connector, account used, date, and person responsible. It seems like bureaucracy until the day someone needs to shut everything off quickly.

The takeaway

ChatGPT Work is useful and the category is here to stay — three of the world's largest software vendors launched the same thing within five months. The mistake is not adopting it; it is adopting it as if it were the same tool as before with one extra button. It stopped being an assistant that answers and became an employee with keys to many doors. Hiring this way requires what any hiring requires: start with minimal access, watch the work closely at the beginning, and expand later.

If your company is at this point — wanting to use it but not knowing where to draw the line — that is exactly what our AI implementation for businesses service is about: deciding what to connect, with which account, with which permission, and in what order, before the agent is inside everything.