OpenAI today released a research preview of Operator, a general-purpose AI agent that can take control of a web browser to autonomously perform tasks like booking travel, making restaurant reservations, and shopping online. The tool is initially available to U.S. users on the $200-per-month ChatGPT Pro plan, accessible via operator.chatgpt.com. OpenAI says it plans to expand to Plus, Team, and Enterprise tiers and eventually integrate Operator into all ChatGPT clients.

Operator is powered by a new Computer-Using Agent (CUA) model that combines GPT-4o’s vision capabilities with advanced reasoning. The CUA interacts with websites through front-end elements — clicking buttons, navigating menus, and filling forms — without needing developer APIs. OpenAI has partnered with DoorDash, eBay, Instacart, Priceline, StubHub, and Uber to ensure compliance with their terms of service.

Operator requires user confirmation before finalizing actions like submitting orders. For sensitive tasks like banking, it demands active supervision, and users must manually enter credit card information. OpenAI warns the CUA is not yet reliable for complex or specialized tasks, and it may get stuck on password fields or CAPTCHAs. Rate limits apply, and Operator currently refuses to send emails or delete calendar events for safety reasons.

Late last week, OpenAI released Tasks, which gave ChatGPT basic automation features akin to Siri or Alexa. Operator represents a far more ambitious step toward the agentic future that Sam Altman has signaled for 2025. The group chats this week are split: some are impressed by the browser-takeover demo, while others wonder whether paying $200 a month for an agent that still needs to ask permission before clicking “buy” is a luxury most can afford.