OpenAI Releases "Agent" in ChatGPT

ChatGPT can now think and take action, using its own computer to proactively choose from a set of agentic tools to complete tasks for you.

OpenAI has launched a new system called ChatGPT Agent. It runs inside a virtual machine and can take actions on its own. That includes browsing websites, running code, accessing tools, and producing outputs like slide decks or spreadsheets.

I haven’t personally tried it since it’s still not fully rolled out to where I live, but based on the demos shared by OpenAI on their YouTube live stream, this is by far the best UX for an agent ever.

The Agent operates independently within ChatGPT. You give it a prompt, and it chooses the tools it needs to complete the task. You do not have to specify which tool to use. The model figures that out on its own.

The entire process runs inside a sandboxed environment that behaves like a virtual desktop, so you can see what it’s doing in real time. It’s pretty neat. It does this by bringing Operator, Deep Research, and ChatGPT into “one unified system.”

Let’s dive deeper into the details of how ChatGPT Agent works and show you some examples.