OpenAI updates Codex: the agent can now take control of your desktop

OpenAI has released a major update to Codex, its agentic programming tool. The new version comes with an expanded set of capabilities, including deeper control over the user's desktop. This lets the agent act far more autonomously on development tasks.

Codex was originally positioned as a tool for automating routine coding work. The new iteration shifts the emphasis: the tool can now do more than generate code snippets. It can also manage the execution environment, interact with desktop interfaces and carry out multi-step tasks without constant human involvement.

The update is clearly a response to competitive pressure from Anthropic. Claude Code, the company's agentic CLI tool, is winning over developers thanks to its terminal integration, support for complex codebases and the ability to run commands directly in the development environment. With the new Codex, OpenAI is going after the same niche.

Desktop control is part of a broader trend toward computer-using agents. Anthropic introduced Computer Use, while Microsoft and Google are actively developing their own solutions in this area. OpenAI is effectively saying that Codex is not just a copilot but a full agent that can work in a developer's real environment.

For the market, this means tougher competition among AI development tools. Companies that have already built their workflows around a single agent now have a real alternative with broader system permissions. The question of trusting agents that have desktop access moves to the forefront.