OpenAI’s ChatGPT gets a general purpose agent

Just a heads up, if you buy something through our links, we may get a small share of the sale. It’s one of the ways we keep the lights on here. Click here for more.

OpenAI is releasing a powerful new feature in ChatGPT called the ChatGPT agent, designed to help users complete a wide range of tasks on their computers. 

Unlike the usual ChatGPT that mostly answers questions, this agent can take action, like navigating your calendar, making slide presentations, running code, and even connecting to apps like Gmail or GitHub to fetch information.

This new agent brings together features from previous OpenAI tools, such as one that could browse websites like a human and another that could research multiple sources to create detailed reports. 

Users don’t need any special skills to use it — just type in plain English what you want done.

To try it out, users with Pro, Plus, or Team subscriptions can switch to “agent mode” inside ChatGPT. 

OpenAI says this is their biggest step yet toward creating AI tools that don’t just respond, but actively do things on your behalf.

ChatGPT can now do work for you using its own computer.

Introducing ChatGPT agent—a unified agentic system combining Operator’s action-taking remote browser, deep research’s web synthesis, and ChatGPT’s conversational strengths. pic.twitter.com/7uN2Nc6nBQ

— OpenAI (@OpenAI) July 17, 2025

The ChatGPT agent is more advanced than earlier versions of AI agents from OpenAI and other tech companies. 

Earlier models struggled with complex tasks, but this new one can, for example, plan a Japanese breakfast for four people, or research business competitors and create a slide deck. 

That requires a lot of behind-the-scenes planning and tool usage by the AI. OpenAI says the agent performs extremely well in tests. 

For example, it scored more than twice as high as earlier models on a difficult exam meant to test broad human knowledge.

It also did well on a very tough math test when given access to tools like a code terminal.

Because the agent can do more, it also poses more risks. OpenAI has added new safety measures. 

It checks prompts in real time for signs of dangerous topics like biology-related threats, and has disabled memory for now to prevent misuse, such as sneaky attempts to steal private data.

While the ChatGPT agent seems very promising, it will take time to see how well it works in everyday use.

Do you think AI agents that can take actions on your behalf are the future of productivity? Or are you concerned about the risks of giving AI so much control over your digital life? Tell us below in the comments, or reach us via our Twitter or Facebook.