What is the new ChatGPT Agent that can ‘control’ your computer?

OpenAI has launched a new software for ChatGPT.

The software, known as ChatGPT Agent, can undertake a wide variety of computer-based tasks for users.

ChatGPT Agent became available on Thursday for subscribers of OpenAI’s Pro, Plus, and Team plans.

Users simply have to choose “agent mode” in ChatGPT’s dropdown menu to activate it.

But what is it? What do we know about it? How does it work? Why is it significant?

What is it?

OpenAI, which is helmed by Sam Altman, claims that ChatGPT Agent goes far beyond being a mere chatbot.

The company says that ChatGPT Agent uses its own virtual computer to ‘think’ and ‘act’.

It essentially functions like a personal assistant to which you can delegate tasks.

This includes executing code, going to websites, managing your calendar, making meal plans, creating presentations and spreadsheets, and summarising meetings.

The company says users can interact with ChatGPT Agent in a ‘natural language’.

The company in its blog said users can issue commands such as “look at my calendar and brief me on upcoming client meetings based on recent news” or “plan and buy ingredients to make Japanese breakfast for four”.

ChatGPT Pro subscribers will be allowed 400 queries per month.

Meanwhile, ChatGPT Team/Plus users will receive 40 queries per month.

It will available to ChatGPT Enterprise and Education users later this year.

Interestingly, the brains behind ChatGPT agent is an Indian.

He is currently in charge of the team behind ChatGPT Agent,

Kumar and Isa Fulford, the research lead on ChatGPT Agent, unveiled the software in a demonstration with The Verge.

Kumar and Fulford in the demonstration asked ChatGPT Agent to plan a date night for a couple.

They also requested that ChatGPT Agent write a research report on Labubus and compare their ascendancy with Beanie Babies.

This includes the web browser tool Operator and Deep Research, an analysis tool which can take information from different websites and write a research report.

Open AI had described Operator as “an agent that can go to the web to perform tasks for you”.

The model behind ChatGPT Agen t, which does not have a name, was trained through reinforcement learning – which is the standard technique for all OpenAI’s reasoning models.

OpenAI says the model delivers state-of-the-art performance on several benchmarks, according to OpenAI.

This includes Humanity’s Last Exam in which it scored 41.6 per cent – which is around double of what OpenAI’s o3 and o4-mini got on the test.

When it comes to the toughest maths tests, OpenAI said ChatGPT Agent hit 27.4 per cent using a terminal that allows it to execute code.

The o4-mini, meanwhile, which was considered the top scorer on FrontierMath, netted a mere 6.3 per cent.

The company combined teams from both Operator and Deep Research to work on ChatGPT Agent.

The team comprised between 20 and 35 people.

Why is it significant?

Because until now, all the AI chatbots have simply sought to answer questions from users.

OpenAI is taking things a step further than its rival s by making ChatGPT Agent more of a personal assistant.

Countries and companies across the world are currently in an AI arms race.

DeepSeek, a previously little-known Chinese firm, shook up Wall Street and Silicon Valley earlier this year.

The Trump administration has vowed to spend $500 billion in the AI race in a project it has dubbed ‘Stargate’.

OpenAI says users can even instruct ChatGPT Agent while the task is unfolding.

“Likewise, ChatGPT itself may proactively seek additional details from you when needed to ensure the task remains aligned with your goals. If a task takes longer than anticipated or feels stuck, you can pause it, ask it for a progress summary, or stop it entirely and receive partial results. If you have the ChatGPT app on your phone, it will send you a notification when it’s done with your task,” says OpenAI in its blog post.

However, the team behind the software warns that ChatGPT Agent is still a bit slow – relatively speaking of course.

“Even if it takes 15 minutes, half an hour, it’s quite a big speed-up compared to how long it would take you to do it,” Fulford told The Verge. “It’s one of those things where you can kick something off in the background and then come back to it.”

Fulford told Wired she asked ChatGPT agent to order cupcakes for her.

“I was very specific about what I wanted, and it was a lot of cupcakes,” she says. “That one took almost an hour—but it was easier than me doing it myself, because I didn’t want to do it.”

For those who worry, ChatGPT Agent also asks for user permission before doing important things such as sending an email or making a booking.

The firm has said that it has built many protections into ChatGPT Agent.

This includes refusing to work on “high-risk” tasks such as bank transfers.

“We have built a lot of safeguards and warnings into it, and broader mitigations than we’ve ever developed before from robust training to system safeguards to user controls, but we can’t anticipate everything. In the spirit of iterative deployment, we are going to warn users heavily and give users freedom to take actions carefully if they want to,” Altman said.

He recommended that users refrain from giving ChatGPT Agent too much personal information.

“We recommend giving agents the minimum access required to complete a task to reduce privacy and security risks,” he added.

Still, Altman took to social media to tout ChatGPT Agent’s potential.

“Agent represents a new level of capability for AI systems and can accomplish some remarkable, complex tasks for you using its own computer. It combines the spirit of Deep Research and Operator, but is more powerful than that may sound—it can think for a long time, use some tools, think some more, take some actions, think some more, etc,” Altman wrote on X.

“For example, we showed a demo in our launch of preparing for a friend’s wedding: buying an outfit, booking travel, choosing a gift, etc. We also showed an example of analysing data and creating a presentation for work,” he added.

With inputs from agencies