Computer-Using Agent

Source: OpenAI Date: 2025-01-23 URL: https://openai.com/index/computer-using-agent

Summary

OpenAI’s launch of CUA — a model capability that allows the AI to operate GUI interfaces directly: clicking, scrolling, typing, and navigating applications as a human would. CUA is built on GPT-4o and represents OpenAI’s entry into the computer-use category that Anthropic opened with Claude’s computer use feature in October 2024. The capability is designed for operator-level task automation without requiring API integrations.

Implications

The agentic computer control thread. CUA is a significant capability shift — from language-in/language-out to language-in/action-out on real computer interfaces. Anthropic shipped this first (October 2024 Claude computer use), forcing OpenAI to follow within months. The race to general computer control is now explicit: whoever wins browser/desktop automation at scale owns the agent substrate for knowledge work.

Integration moat vs. GUI scraping. CUA’s value proposition is that it doesn’t need structured APIs — it can use any software humans use. This is both the power and the risk: it bypasses integration layers that would normally create developer relationships. Watch how enterprises evaluate CUA vs. purpose-built RPA (UiPath, Automation Anywhere) and whether OpenAI’s Codex CLI + CUA combination becomes the preferred operator stack by mid-2025.