Introducing ChatGPT agent
OpenAI ● Covered by 2 sources
OpenAI just gave ChatGPT hands. It can now browse, click, and use tools on its own to actually finish tasks, not just chat about them.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has rolled out something it calls ChatGPT agent, and the pitch is simple: stop asking the bot for advice and start letting it do the thing. Research a topic, book a flight, build a slideshow — the agent handles the multi-step grind that used to mean twenty browser tabs and an afternoon lost.
What's actually new here isn't that ChatGPT can write good prose or summarize a PDF. It's that it can now act inside tools with a human still holding the leash. You give it a goal, it plans a sequence of steps, executes them using whatever software or web interface the task requires, and checks back with you along the way. Think less chatbot, more junior assistant who actually goes and does the errand instead of just telling you how.
OpenAI is framing this as the next phase of ChatGPT's evolution, moving from a text generator to something closer to an operator. That's a meaningful shift in ambition. Booking things and assembling slideshows are mundane examples on purpose — they're the kind of chores that don't require genius, just follow-through, and follow-through is exactly what earlier chatbots were bad at.
The guidance piece matters too. This isn't OpenAI handing ChatGPT the keys and walking away. The agent is designed to work with your oversight, presumably because letting an AI wander around the internet unsupervised making purchases is the kind of thing that ends up in a very different kind of blog post. Expect OpenAI to lean hard on that human-in-the-loop framing as it rolls this out more broadly, especially given how nervous people already are about AI systems that can take real-world actions instead of just suggesting them.
My take — AI-written commentary, not fact-checked reporting
This is the move I've been waiting for, honestly — chatbots that talk a good game but can't lift a finger were always half a product. But 'with your guidance' is doing a lot of quiet work in that sentence, and I'd bet real money the interesting failures show up the first time someone lets this thing book something without watching closely.
Read more about this at: OpenAI