OpenAI Updated Codex: Agent Now Takes Control of Desktop

OpenAI has released a major update to Codex, its agentic tool for programming. The updated version has gained an expanded set of capabilities, including deeper control over the user's desktop. This allows the agent to act significantly more autonomously when solving development tasks.

Codex was initially positioned as a tool for automating routine code tasks. However, in the new iteration, the emphasis has shifted: the tool is now capable of not just generating code snippets, but also managing the execution environment, interacting with desktop interfaces, and performing multi-step tasks without constant human involvement. The update is clearly aimed at competitive pressure from Anthropic.

Claude Code — the company's agentic CLI tool — is actively gaining audience among developers thanks to integration with the terminal, support for complex code bases, and the ability to run commands directly in the development environment. With the new Codex, OpenAI is competing for the same niche. The ability to control the desktop is part of a broader trend toward computer agents.

After Anthropic introduced Computer Use, and Microsoft and Google are actively developing their own solutions in this area, OpenAI is essentially stating: Codex is not just a copilot, but a full-fledged agent capable of working in a real developer environment. For the market, this means intensifying competition in the AI tools for development segment. Companies that have already built workflows around one agent now have a real alternative with broader system rights.

The question of trust in agents with desktop access comes to the forefront.