OpenAI Launches ChatGPT Agent to Automate Complex Tasks

Last year, OpenAI launched Operator, a tool that could perform basic tasks for you. Now they’ve taken it a step further with the new ChatGPT Agent. Instead of just acting, it reasons through each step before completing it. It also connects to your apps like Calendar, Gmail, and Drive to get things done for you. Think of it as Operator, Deep Research, and Connectors all rolled into one.

Let’s break down what this new ChatGPT Agent is, what it can do, how it works, and who can use it.

What is ChatGPT Agent?

ChatGPT Agent is a new tool inside ChatGPT that uses a virtual computer to perform real, online tasks on your behalf. Instead of just replying with text, it can:

• Visit websites
• Click buttons and fill out forms
• Look through your Google Calendar
• Pull files from Google Drive or Gmail
• Generate slides or spreadsheets
• Analyze information and write detailed reports

In short: you give it a job, and it does the clicking, searching, reading, and writing for you.

Real-World Examples

OpenAI showed off a bunch of use cases to prove how useful this can be:

• Look at your calendar and give a meeting brief based on recent news
• Plan a Japanese breakfast, order ingredients online
• Compare competitors and make a PowerPoint
• Plan a date night by checking your schedule and restaurant availability on OpenTable
• Request office parking every week automatically

You can even interrupt it midway and add new instructions, like changing the restaurant type while it’s still working.

What Powers ChatGPT Agent?

This new ChatGPT Agent is built on a fresh model trained specifically for complex, multi-step tasks. It merges three powerful pieces:

• Operator – A visual browser that can click around the web like a human
• Deep Research – A text-based research agent that can browse, summarize, and cite information
• Connectors – Tools that let it connect to Gmail, Drive, Dropbox, and other data sources

Together, they give the agent a virtual workspace: a sandboxed terminal, file system, and browser that only it can use.

It can pick up a file from Drive, read its content visually, analyze it deeply, and build a PDF, spreadsheet, or presentation based on what it finds.

It’s Not Fast—but It’s Smart

Some tasks take 15 to 30 minutes. But that’s the trade-off: this isn’t about instant answers. It’s about offloading real work. Let it run in the background while you do something else.

OpenAI says this is like having a digital assistant who actually works. You don’t need to babysit it.

What About Privacy and Safety?

It won’t do anything serious—like send an email or make a payment—without asking you first.

There’s also a Watch Mode: if you ask it to do something on a financial site, you can’t switch tabs or leave the page, or it stops working. That prevents misuse.

And even though there’s no proof it can help someone create something dangerous, OpenAI has already activated safeguards designed for high-risk use cases like biology or chemistry.

How to Use ChatGPT Agent?

If you’re a ChatGPT Pro user, you should see “Agent Mode” starting today under the Tools menu. Or just type /agent in the chat to enable it. Plus and Team users will see it roll out in the next few days, while Enterprise and Education users will get access later this month.

• Pro users: 400 agent messages/month
• Plus & Team users: 40/month
• Enterprise and Education: Coming soon

It’s not yet available in the European Economic Area or Switzerland.

The Bigger Picture

ChatGPT Agent is part of a bigger AI shift. Companies like Klarna, Anthropic, and Google are all building agents that act more like employees than tools. OpenAI’s goal is to move closer to a J.A.R.V.I.S.-like assistant—one that can plan, think, act, and adjust.

This is one big step in that direction. And for the first time, ChatGPT isn’t just a chatbot. It’s a worker.