OpenAI's new AI agent can browse web for you
What's the story
OpenAI has unveiled a research preview of Operator, an advanced artificial intelligence (AI) agent, which can perform tasks on a web browser on its own.
From filling out forms to ordering groceries and even creating memes, the range of things this groundbreaking tool can do is pretty impressive.
Operator is available to US users on ChatGPT's $200 Pro subscription plan at the moment.
Expansion strategy
Global rollout and integration plans
OpenAI CEO Sam Altman has confirmed that Operator will be rolled out to other countries soon, but Europe may have to wait a bit longer.
The company also plans to integrate Operator into all of its ChatGPT clients in the future including Plus, Team, and Enterprise.
The initial research preview of this groundbreaking AI agent is currently accessible via operator.chatgpt.com.
Functionalities
Task automation capabilities and user interface
Operator has been designed to automate tasks like booking travel accommodations, making restaurant reservations, and online shopping.
Users can pick from a number of task categories on the Operator interface, including shopping, delivery, dining, and travel, all of which enable different kinds of automation.
When activated by ChatGPT users, Operator opens a dedicated web browser to complete tasks while explaining what exactly it is doing.
Tech collaboration
Unique technology and collaboration with businesses
Operator is powered by a new computer-using agent (CUA) model that merges the vision capabilities of OpenAI's GPT-4o model with reasoning capabilities from its more advanced models.
The CUA interacts with websites like a human, using buttons, navigating menus, and filling out forms on a web page much like a human.
OpenAI is working with companies like DoorDash, Instacart, Priceline, StubHub, and Uber to ensure Operator complies with their terms of service agreements.
Performance reliability
Performance reliability and user confirmation feature
OpenAI has said that the CUA model is trained to ask for user confirmation before finalizing tasks with external side effects. This way, the users get a chance to review the model's work before it goes permanent.
However, OpenAI also warns that CUA may not be reliable in all scenarios just yet.
In case of issues or errors, Operator can reason itself out or hand control back to the user if required.