In past couple of years, ChatGPT has made its way into your workflows as your “thinking partner.” It thinks on your behalf, corrects where you’re wrong, helps you write content for presentations, and even draft corporate-sounding emails to send to your colleagues.

And during this time, we’ve seen ChatGPT’s thinking capacity improve dramatically. Its ability to conduct web searches, assist with code, handle complex questions, and even generate images has become more sophisticated, making it even more useful.

But now, there’s a new chapter. ChatGPT is all ready to act on your behalf as well.

OpenAI launches a general purpose agent

On July 17, 2025, OpenAI announced the launch of a ChatGPT agent – an update that will go beyond conversations and ideas. This general purpose agent is meant to autonomously carry out tasks for you. Whether it’s browsing the web, filling out forms, running code, creating presentations, or even making purchases, ChatGPT Agent can now handle these actions directly.

But let’s take a step back and think about what ChatGPT used to be and what it can now achieve.

Before:

ChatGPT would help you draft an event agenda or write an email. But you’d still need to handle the execution yourself.

Now:

ChatGPT agent can act. It can  drafts the agenda, sends invitations, books the venue, and handles every detail.

Now, let’s break it down.

Traditional models like ChatGPT, which are based on large language models (LLMs), are designed to process and generate text. They excel at offering solutions, ideas, and information, but they can’t perform actions beyond providing responses. Their role is limited to thinking and suggesting.

AI agents, on the other hand, combine thinking and doing. They not only provide solutions but also act on them.

For example: see how this user ordered his dinner with the ChatGPT agent.

BREAKING: OpenAI just released ChatGPT agent.

I just tried it and I'm blown away.

(It did a tesco food shop for me and ordered a roast dinner with sticky toffee pudding): pic.twitter.com/nBSxy5YWNq

— Alex Banks (@thealexbanks) July 17, 2025

👉 For a deeper understanding of AI agents and how they differ from traditional language models, check out our guide on AI agents.

What can the ChatGPT agent actually do?

ChatGPT Agent can now handle a range of tasks that were once up to you. Here’s what it can do:

• Web Actions: Browse websites, complete forms, and interact with online services.
• Run Code: Execute code to solve problems or automate tasks.
• App Integration: Connect with apps like Gmail, Google Docs, and more.
• Create Content: Automatically generate slides, spreadsheets, and other documents.

In short, ChatGPT Agent takes your ideas and turns them into action, making tasks faster and easier than ever.

How the ChatGPT agent actually works?

Behind the scenes, ChatGPT Agent combines the strengths of two powerful tools: Operator and Deep Research.

Operator allows ChatGPT Agent to interact with websites, fill out forms, and perform tasks that require browsing. Deep Research enables it to gather and analyze information from multiple sources to provide comprehensive insights.

Previously, users sometimes provided queries that were better suited for one tool over the other. This separation led to fragmented workflows and increased complexity.

To address this, OpenAI unified these capabilities into ChatGPT Agent. Additionally, ChatGPT Agent leverages ChatGPT’s conversational intelligence, ensuring that interactions remain natural and intuitive.

And don’t worry — you’re always in control. ChatGPT Agent will ask for your permission before taking significant actions, and you can interrupt or stop tasks at any time. This ensures that you have full oversight and can guide the process as needed.

How are AI agents changing the game?

ChatGPT Agent isn’t the only AI assistant evolving to handle tasks beyond conversation. Companies like Google and Anthropic are also developing similar capabilities.

Google’s Project Mariner, for instance, allows AI to navigate websites, fill out forms, and complete online tasks autonomously. Anthropic has introduced Claude Opus 4 and Sonnet 4, models designed to perform complex tasks and integrate with various applications.

As these AI agents become more capable, there’s speculation about how companies might monetize their use. Some believe that if an AI agent assists a user in completing a purchase, the company could take a small percentage of the transaction.

In past, OpenAI’s CEO, Sam Altman, had indicated that they plan to take 2% of sales generated through their Deep Research tool.

This evolving landscape suggests that AI agents could not only enhance productivity but also become integral to new business models in the tech industry.

ChatGPT Agent is now available to Pro, Plus and Team users outside of the EU area.