OpenAI Launches Operator, an AI Agent Capable of Browsing and Performing Internet Tasks

OpenAI has presented Operator, an artificial intelligence agent designed to perform repetitive tasks in web browsers autonomously. This new technology represents a significant leap in AI systems' ability to interact with digital interfaces.

The new system, based on the Computer-Using Agent (CUA) model, can perform actions such as filling out forms, making online purchases, creating memes, and managing various web tasks without requiring custom API integrations. The tool uses GPT-4o's vision capabilities and advanced reinforcement learning techniques to navigate and interact with graphical user elements.

Initially, Operator will be available for Pro users in the United States, with plans to expand to Plus, Team, and Enterprise users. The company has developed this technology with a cautious approach, implementing multiple security layers to ensure responsible use.

Among the most notable features are the ability to self-correct, the possibility of requesting human intervention in complex tasks, and workflow customization options. OpenAI has collaborated with companies like DoorDash, Instacart, and Uber to ensure the tool responds to real needs.

The platform includes important safeguards, such as supervision mode, user confirmations for significant actions, and limitations on sensitive tasks. Additionally, it offers privacy options like browsing data deletion and the ability to opt out of model training.

Despite its promising capabilities, OpenAI acknowledges that Operator is in a research preview phase, with known limitations and significant room for improvement based on user feedback.