OpenAI Releases AI Agent "Operator"

The startup OpenAI has introduced its own AI agent called ”Operator.” It is capable of performing tasks on the internet on behalf of the user. The new tool can browse web pages and interact with them, type text, scroll, and click buttons.

Operator can be asked to perform a variety of repetitive basic tasks such as filling out forms, ordering groceries, or booking hotels. ”The ability to use the same interfaces and tools that people interact with daily expands the scope of AI application, helping to save time on everyday tasks and opening up new opportunities for business interaction,” OpenAI's announcement states.

Operator is powered by a new AI model called Computer-Using Agent (CUA). It combines GPT-4o's ability to see the screen with advanced reasoning through reinforcement learning. The agent perceives information through screenshots and can perform the same actions as a human using a mouse and keyboard.

The model is trained to request confirmation before completing tasks such as booking a hotel or sending an email.

Currently, a preliminary research version is functioning and will be developed based on user feedback. The AI agent is available to ChatGPT Pro subscribers for $200 in the US on a special resource. Future plans include expanding access to more people.

At this stage, the agent doesn't work perfectly, and if it encounters difficulties, it will ask the user to complete the task.

As a reminder, in October 2024, the AI startup Anthropic released an updated version of the Claude 3.5 Sonnet model, which can interact with a computer like a human - moving the cursor, clicking buttons, and typing text.