OpenAI just announced ChatGPT Agent – here's everything you need to know as it happened

ChatGPT Agent is rolling out today for Pro, Plus, and Team users

Well, that livestream covered a lot of ground and gave us a quick deep dive into ChatGPT Agent.

Here's the important detail, though. ChatGPT Agent is going live today for Pro, Team, and Plus users. Folks on the Pro plan will get 400 queries a month with ChatGPT Agent, while Team and Plus users get 40 a month. OpenAI expects the rollout to be completed 'soon' for Team and Plus, while all Pro users should receive it today.

Sam Altman closed, noting the risks and forthcoming ones, again, AI Agents. Noting that ChatGPT Agent will feature a robust alert system, with lots of warnings everywhere, but that will relax over time.

A serious upgrade for ChatGPT

Now, ChatGPT Agent appears to be a pretty major upgrade to ChatGPT, and from the demos thus far, it could be a handy tool, especially for individuals already engaging with the AI in some of these ways.

It also expands the capabilities of ChatGPT Operator and pairs it with more advanced models housed within Deep Research. Arguably, placing both in an easier format to work with.

The OpenAI team is highlighting one area of concern with ChatGPT Agent, specifically whether it could be compromised by a malicious website on the internet. That's especially important considering that it could have your account information or banking credentials to make a purchase.

OpenAI notes that using takeover mode would be ideal in these scenarios, but also not oversharing or being critical around what you're asking ChatGPT Agent to do. Sam Altman said a 'new level of attacks' would come towards it.

This one is a bit more advanced – asking ChatGPT Agent to retrieve evaluation numbers from Google Drive and then compile them into a slide presentation.

You can interrupt ChatGPT Agent

This is pretty neat – you can also interrupt ChatGPT Agent while it's in the middle of a task. The OpenAI team shared that this was a key capability for the Agent to be able to deliver, and while it might take a second to respond, as we saw moments ago, this is helpful, especially for larger tasks.

The model is also trained to ask for confirmation before completing the last steps. This could be to catch a typo in an email or some other detail that you really need to get right. The primary goal is for ChatGPT Agent to be highly collaborative.

And yes, in the background of the event, ChatGPT Agent is still planning the wedding. One neat feature of Agent is that you can scroll back to see what it's been doing and what it has accomplished so far.

While this is still being worked on regarding the wedding request, the OpenAI team is switching to ChatGPT on an iPhone to ask the Agent to create some stickers of the team mascot.

OpenAI is reminding us that Operator, an agent for web browsing, and Deep Research both shipped separately earlier this year, but the team realized that the tools are complementary. Both of these capabilities are integrated within ChatGPT Agent, and it's also trained to leverage other tools.

ChatGPT Agent is official

Well, Sam Altman – OpenAI CEO – is excited as he kicked off the livestream by saying that 'we have a banger today' and quickly confirmed ChatGPT Agent. The entire concept is a unified agent that can handle the legwork, make informed decisions about which models to use, and navigate the web independently – that's this new Agent, and it brings several key elements together.

And OpenAI is wasting no time, firing up ChatGPT on the web and clicking 'Agent' right from the options to use the new tool. The quest here is to help with attending a wedding – finding a gift and even hotels. The team says that Agent is using a computer, setting up an environment, and then goes about understanding the prompt. It may even ask for more information or a clarifying question.

The ChatGPT Agent has access to its own computer of sorts, and ChatGPT will actually show you a visualization of what the Agent is doing. OpenAI has given the Agent access to a number of tools, including a terminal to write code and other ChatGPT APIs.

OpenAI also used reinforcement learning to train the Agent on how to pick a tool to use, depending on the prompt. This is really, really neat.

We're just four minutes away from the OpenAI team kicking things off, and along with some smooth, peaceful tones playing, we're seeing a screen that says "ChatGPT agent Starting Soon" with a blue cursor swimming around.

10 minutes to go!

We're just 10 minutes away from OpenAI's July 17, 2025, event kicking off, and along with the description for the livestream on YouTube, the title is also giving some more information away: Introduction to ChatGPT agent.

We'll be breaking it down as OpenAI makes its announcements, and you can watch the livestream right above.

Live stream description offers some clues...

The YouTube page for the live stream of the OpenAI announcement has offered some clues as to what we might be about to get.

In the description for the live stream it says: "Join Sam Altman, Casey Chu, Isa Fulford, Yash Kumar, and Zhiqing Sun as they introduce our unified agentic model in ChatGPT."

So, what does a "unified agent model in ChatGPT" mean? It could be a new ChatGPT model that will decide for you which tool would best suit your query. So, instead of you selecting more powerful LLM it would decide if what you're asking is complex enough to deserve a more powerful reasoning model, or deep research tool, or if you would be better off with a more lightweight model.

Interesting times!

What will OpenAI release today?

There’s a lot of speculation about what OpenAI will be announcing today, after a mysterious video on its X.com feed:

However, an hour ago it posted again with this update:

Which contains the words: “ChatGPT / Deep research / Operator. Livestream in 3 hours.” There are handshake icons between each of the three words, indicating that the systems are talking to each other in some way.

So what happens when ChatGPT talks to Deep research, which talks to Operator? Some have speculated that OpenAI are about to release a new browser with agentic properties.

The timing would be interesting, since the new Perplexity browser called Comet has just been released and we were certainly impressed with our first look. John-Anthony Disotto commented: “In an age where we are constantly seeking innovations to transform our lives, it’s become increasingly common for products and software to not live up to their billing. Comet might still have its quirks, like issues related to some prompts crashing when the AI meets a hurdle, but initial impressions leave me yearning for more. I didn’t understand the point of an AI web browser, but after 48 hours with Comet, I’m convinced that having a smart assistant built into your internet experience will transform our lives forever.”

Let's kick things off with the teasers that OpenAI posted on X this AM, and a reminder we're about 45 minutes away from the livestream starting.