Operator: the AI agent that browses the web and performs tasks for you
OpenAI unveils Operator, an AI agent capable of ordering online and browsing the web
OpenAI, the company behind ChatGPT, recently presented Operator, a new AI agent that can navigate the web and perform tasks for users. This technology is revolutionizing the online experience by automating activities like booking tickets or ordering rides, with minimal human intervention.
What is Operator and how does it work?
Operator is based on the model Computer Using Agent (CUA) from OpenAI, which combines the visual capabilities of GPT-4o with advanced reasoning methods. This allows the agent to browse websites, fill out forms, click buttons, and scroll through pages, as a human user would.
According to Yash Kumar, product and engineering manager at OpenAI, “He can navigate websites and perform actions on them, just like you and me.” Unlike traditional AI tools that use APIs, Operator leverages screenshots and a virtual browser to visually analyze the elements, such as buttons and text fields.
A project in development: limits and security concerns
While Operator marks a major step forward, OpenAI acknowledges that it is still in its early stages and has some limitations. Complex tasks, such as creating slideshows or organizing detailed events, remain difficult to automate. In demonstrations, Operator displayed a 87% success rate for online tasks, but had difficulty with multi-step processes.
La security and confidentiality are at the heart of OpenAI's deployment strategy. The agent uses a three-level security system to ensure maximum user control:
- Support mode : for sensitive actions (e.g. password entry, payments), Operator requires manual intervention.
- User confirmations : before any major action (e.g. validation of a purchase), the user must approve the operation.
- Task Restrictions : Operator refuses high-risk actions, such as financial transactions or job applications.
OpenAI has also implemented a monitoring mode to prevent abuse and phishing attacks.
Partnerships and concrete applications
OpenAI collaborates with DoorDash, Instacart, Uber and OpenTable to allow Operator to handle practical tasks, such as ordering meals or booking services. The company also works with the city of Stockton to facilitate access to public services.
According to Jamil Niazi, Stockton's director of information technology, “By learning how to better use Operator, we will be able to identify how AI can simplify civic engagement for our residents.”
A glimpse into the future of AI agents
Sam Altman, CEO of OpenAI, introduced Operator as the first in a series of AI agents in development. In the future, these agents will be integrated with ChatGPT and an API will be made available for developers to create their own AI solutions.
Availability and outlook
Currently, Operator is reserved for ChatGPT Pro subscribers in the United States, with a monthly cost of 200 dollars. OpenAI plans a global launch after improvements and user feedback.
Although promising, Operator still requires human supervision. Its launch, however, marks a transition towards active AI agents, capable of handling both online purchases and daily planning.
To go further: how AI agents are transforming the workforce, What is an AI agent? et the era of GPT agents.