In a quick live stream today, OpenAI announced ChatGPT agent, which improves on its promising, but limited Operator feature announced a few months ago. Here’s how it works
Operator meets Deep Research
What OpenAI did was combine Operator, which already had agentic browsing capabilities, and Deep Research, which can quickly consume, understand, and infer on website content.
The result is an agent that gets spun up on a virtual machine when you assign it a task, and then it just… goes.
It can ask follow-up questions to clarify the user’s intent, and can ask the user to log in to services in order to complete transactions.
Users can take over at any point, as well as send in complementary instructions halfway through.
The way OpenAI set up ChatGPT agent is pretty interesting. It basically combines:
Text-based browsing through Deep Research
through Deep Research Visual browsing through Operator
... continue reading