OpenAI Unveils Operator, Its First Autonomous AI Agent for Web Tasks
OpenAI releases Operator, a Level 3 AI agent that autonomously completes browser-based tasks like shopping and booking. The tool is currently available exclusively to U.S. ChatGPT Pro users.
OpenAI launches Operator, marking its first major step into autonomous AI agents under its AGI roadmap. Unlike traditional chat-based assistants, this new tool actively executes real-world tasks such as booking reservations, ordering groceries, and scheduling appointments directly within a web browser. The service debuts exclusively in the United States for ChatGPT Pro subscribers.
The system relies on a new CUA architecture powered by GPT-4o that combines vision and reasoning to navigate the web. It analyzes screen pixels to understand visual layouts, simulates mouse clicks and keyboard inputs, and connects multiple steps into seamless workflows. Operator already partners with major platforms like Instacart, OpenTable, and Uber, while featuring built-in error recovery that pauses for user approval during critical actions.
Despite its groundbreaking capabilities, Operator currently faces notable limitations. It achieves a 58 percent success rate compared to 78 percent for humans, struggles with highly complex or custom interfaces, and restricts high-risk actions like banking. OpenAI plans to address these challenges by releasing a developer API in the third quarter of 2025 and expanding the service globally to Europe and Asia later in the year.