Operator — also known as CUA, short for "Computer Using Agent" — is based on OpenAI's newest GPT-4o vision model. It is an agent system that controls its own web browser and can be programmed to scan the elements on the browser's screen and take actions: fill out a SEARCH field, choose an item from a drop-down menu, click a SUBMIT button, and so on.

It seems that Operator has everything needed to make its 'old' sibling, RPA systems, a thing of the past.

• Availability (Feb 2025) — USA only
• Subscription required — OpenAI Pro plan

Traditional RPA systems rely on brittle, script-based automation: they follow predefined rules and break the moment a UI changes. Operator, by contrast, uses computer vision and reasoning to understand what it sees on the screen, making it far more resilient to interface changes.

Where a classic RPA bot needs explicit coordinates or element selectors, Operator can read a page much like a human would — interpreting labels, buttons, and forms in context — and decide what action to take next.

• No more brittle selectors — Operator adapts to UI changes without needing script updates.
• Vision-driven interaction — It "sees" the browser just as a human does, using GPT-4o's vision capabilities.
• Agentic by design — It can plan a multi-step task autonomously.
• Broad use cases — From e-commerce checkout flows to enterprise portals, any web-based workflow becomes a candidate for automation.

If Operator or a similar CUA matures quickly, the $15B+ RPA market faces serious disruption — particularly for web-browser-centric use cases.

• Available only in the USA at launch
• Requires an OpenAI Pro subscription
• Still in early rollout — edge cases, complex flows, and CAPTCHAs can trip it up
• Privacy and security considerations for tasks involving sensitive data

This is one to watch closely. The pace at which OpenAI iterates means today's limitations may well be solved features by Q3 2025.