OpenAI Updates Codex to AI Agent That Controls Desktop, Automating Development Workflows

Gate News message, April 17 — OpenAI has upgraded Codex, transforming it from a coding assistant tool into an agent-based service that runs directly on users’ computers and can control applications autonomously. The new version operates in the background, capable of opening and manipulating apps, automating code writing, modification, testing, and review processes.

The updated Codex supports task continuity, allowing it to resume interrupted work and handle long-term development projects consistently. It can review GitHub feedback, manage remote environments, and operate through a UI-based interface where developers can view changes on an in-app browser and direct modifications. The tool also integrates with over 100 applications including Slack, Gmail, and Notion, enabling it to handle workflows beyond development, such as task organization and follow-up suggestions.

The upgrade represents a strategic shift from a terminal-based coding tool to an AI that manages entire desktop environments. OpenAI is also providing a separate agent SDK and integrating the technology into enterprise workflows. Competitor Anthropic has released similar capabilities through Claude and Coworker, automating computer control while users are away. Industry analysts note that AI coding tool competition is rapidly moving from simple code generation to comprehensive workflow automation agents.