OpenAI’s computer-controlling AI agent has arrived — here’s what it can do

OpenAI has introduced “Operator,” a new AI agent that can autonomously perform web-based tasks for ChatGPT Pro subscribers in the United States.

Core technology and capabilities: Operator employs a “Computer-Using Agent” (CUA) model that combines GPT-4o’s visual processing abilities with reinforcement learning to interact with web interfaces like a human user.

• The agent can interpret screenshots and perform basic computer actions like typing, clicking, and scrolling
• Operator navigates web interfaces independently to complete tasks such as ordering groceries and making reservations
• Unlike traditional AI models, Operator doesn’t rely on predefined APIs, allowing for more flexible interaction with websites

Strategic partnerships: OpenAI has formed collaborations with major digital service providers to ensure seamless integration and compliance with terms of service.

• Key partners include DoorDash, Instacart, OpenTable, Priceline, StubHub, Thumbtack, and Uber
• These partnerships help refine Operator’s functionality and expand its practical applications
• The collaborative approach ensures the agent works within established business frameworks

Safety and control measures: OpenAI has implemented robust safety protocols to maintain user control and protect sensitive information.

• The system automatically transfers control back to users when encountering difficulties
• User approval is required for critical actions like sending emails
• Operator requests user intervention when handling sensitive information such as login credentials

Current limitations and availability: The technology is currently in a controlled release phase with some functional constraints.

• Access is limited to ChatGPT Pro subscribers in the United States
• The agent may struggle with complex interfaces like slideshow creation and calendar management
• OpenAI plans to expand access across more user tiers and integrate Operator into ChatGPT

Looking ahead: While Operator represents a significant advancement in autonomous AI agents, its controlled rollout and current limitations suggest a cautious approach to deployment, balancing innovation with practical considerations about safety and reliability.