What is ChatGPT agent mode, and what replaced it?

ChatGPT agent mode was a feature that let ChatGPT carry out tasks for you, such as researching, filling in web forms, and building a shopping list, instead of only replying with text. OpenAI retired it in August 2026. The same idea now lives in ChatGPT Work, which handles longer, multi-step jobs and asks before important actions.

In this guide, you will learn what an agent is, what happened to agent mode, and how to use its replacement sensibly. If you are brand new to ChatGPT, start with our guide on how to use ChatGPT.

What is an AI agent, in plain terms?

A normal chatbot is like a very knowledgeable friend on the phone: it tells you what to do, and you go and do it. An agent is more like an assistant sitting at a computer for you. You describe the goal, and it takes the steps itself: opening websites, clicking buttons, typing into forms, and putting results together.

That shift from answering to acting is what people mean by an AI agent. Our guide on what AI agents are explains the idea in more depth.

What did ChatGPT agent mode do, and why did it go?

Agent mode gave ChatGPT its own browser running on a separate computer. It could read pages, click through them, and do things like research a meal plan and add the ingredients to an online shopping basket ready for you to check. It paused to ask for confirmation before important steps.

In August 2026, OpenAI removed agent mode from ChatGPT. Its help page now directs people to ChatGPT Work instead. You may still find older articles and videos describing agent mode, so do not worry if the steps they show no longer match your screen.

What is ChatGPT Work?

ChatGPT Work is OpenAI’s current agent. It is designed for bigger jobs that take several steps, such as researching a topic, analysing information, or producing a document, spreadsheet, presentation, or report. It can work on a task for a long time, breaking it into smaller steps.

For tasks on websites, Work can use a cloud browser, meaning a web browser running on OpenAI’s computers rather than yours. It can read pages, click buttons, and fill in forms, and it pauses when it needs your input, a sign-in, or your confirmation.

OpenAI says Work is available on every plan in the ChatGPT desktop app, and on the website and phone apps for paid plans. Features roll out gradually, so what you see may differ.

How do you use ChatGPT Work?

1. Open ChatGPT and switch to Work. In the desktop app, OpenAI describes a toggle at the top of the page for choosing Work. On the website and phone apps, look for the Work option if your plan includes it.
2. Describe one clear goal. Say what you want at the end, who it is for, and any limits such as budget or length.
3. Connect only what the task needs. If it asks to use an app such as your email, allow only the ones this job requires.
4. Watch the first few steps. Check it has understood you, and stop or redirect it if it heads the wrong way.
5. Review before you confirm. When it asks permission to book, pay, or send something, read the details carefully first.

Here is a good first task to try:

Research three family-friendly day trips within an hour's drive of
Leeds for a rainy Saturday in October. For each, give opening hours,
rough cost for two adults and two children, and one downside. Put it
in a short table. Do not book or buy anything.

That last line matters: saying what it must not do is as useful as saying what it should.

What is it good for, and what is it not good for?

Good for	Not good for
Research that means visiting many websites	Anything you cannot afford to get wrong
Turning research into a document or slides	Tasks where you cannot check the result
Comparing options and summarising them	Vague requests like “sort out my inbox”
Repetitive form-filling you can review	Handling passwords or card details for you

Agents usually take longer than a normal chat and can still make confident mistakes. Treat the result as work from a capable but new assistant: worth checking.

How do you keep yourself safe?

Handing over actions means handing over some control, so a few habits help:

• Never type passwords or card numbers into the chat. When a site needs a sign-in, ChatGPT shows you the login screen so you can enter your own details.
• Be sparing with connected apps. Every app you connect is something the agent could read or act on.
• Know that websites can try to trick it. A page can hide instructions aimed at the AI, an attack called prompt injection. This is one reason OpenAI makes Work ask before consequential actions like payments.
• Check your app permission settings. Some settings let low-risk actions run automatically. Only loosen them once you trust how it behaves.

Used this way, an agent can save you real time on research and admin while you stay firmly in charge of anything that matters.