ChatGPT agent is the next evolution of AI
Everything you need to know about the the tool that can do work for you on its own computer
Today, OpenAI announced ChatGPT agent, a tool that can do work for you on its own computer.
Here are some things the agent can do:
Reason (like the o-series models and DeepSeek)
Use the web (including clicking buttons and filling out forms)
Write and run code (the run part is important)
Connect to apps (like Gmail and Calendar)
Creating beautiful slideshows
Create valuable spreadsheets
Personally I’m most excited for the Calendar integration. It feels long overdue to have a natural language interface for my calendar. Today’s announcement was pretty exciting for me, so let’s dig into why.
Why ChatGPT agent is a BIG deal
3 words.
Humanity’s last exam.
Humanity's Last Exam is one of most challenging benchmarks for LLMs. It was specifically developed in response to the popular AI benchmarks having reached "saturation" (meaning models were all scoring so high, so the benchmark become not useful).
ChatGPT agent scores twice as high as o3.
This is nuts.
Before today, Operator could scroll, click, and type on the web, but it wasn’t very smart or useful. Meanwhile, deep research excelled at analyzing and summarizing information. Today’s announcment provides a product that excels at both tasks.
Agent excels at spreadsheets and even novel math problems.
To give you a better feel for how useful this is announced to be, here are a few prompts featured in the announcment:
Intelligently updating spreadsheets
Update the attached sheet with 2024 earnings report numbers for ACME Corp, separating each quarter into its own tab and using the same formatting.
Searching for flights
Find me a flight to NYC. I have status with United.
Designing and purchasing stickers
Create an anime sticker of my company’s logo, then order 500 and send them to <address>
When is ChatGPT agent coming out?
Agent rolled out today (July 17, 2025) for Pro customers. It’s rolling out Plus (that’s me!) and Team users on July 21st. Education users can expect it a bit later.
Historically, OpenAI has some problems keeping their infrastructure online to meet demand when they do big releases like this. Fingers crossed they scale okay this time 🤞
So how do I use ChatGPT agent?
Well it’s hard to say since it’s just now rolling out, but I promise to write more on using the agent to do meaningful knowledge work as soon as it’s rolled out to Plus users. OpenAI does say that you’ll be able to select agent from teh dropdown just like deep research, but at any point in the conversation.
One thing that’s become clear in the last year is that OpenAI is increasingly a product company rather than a model company. That may sound like I’m unimpressed with the models, which isn’t true. Instead, I’m even more impressed by their ability to apply AI in a way that is easy to use and provides clear value. So I’m expecting that the agent will be remarkably intuitive.
That being said, I write The AI Augmented Engineer to teach people like you how to use AI at work. Most of my readers are software developers or engineering managers, so most of the content is focused on how to use AI to build better software. If that’s you, consider subscribing!
If you’re new to the newsletter, you might enjoy starting on this page, which outlines some of the most popular posts:
I'm excited to try it on the 21! I upgraded to Pro for Operator a while back but it was useless so I got a refund and downgraded.
that sounds really exciting. But any how to build an agent for a new beginner in term of AI Agent？