OpenAI's Codex Gains Desktop Control, Posing New Challenge to Claude

OpenAI is significantly expanding the capabilities of its Codex development system, introducing features that allow it to directly interact with a user's computer, create images, and retain information across sessions. This move signals a clear escalation in the competitive field of AI-assisted software development.

The most notable addition grants Codex the ability to operate macOS applications autonomously. According to OpenAI, the system can work in the background without disrupting a user's primary workflow, with multiple agents capable of running simultaneously. The company suggests this will aid developers in testing frontend modifications, evaluating applications, or working with software that lacks a public API. This functionality launches first for macOS users of the Codex desktop app, with plans for other operating systems and the European Union to follow.

Beyond desktop control, the update brings several other enhancements. Codex can now generate and refine images using a new model, integrates with tools including GitLab and the Microsoft Suite, and features a native web browser for direct page interaction. Task automation is also streamlined, with options to reuse conversation threads and schedule future work.

A new memory feature, offered as an opt-in preview, will let Codex store useful context from previous interactions—such as user preferences or hard-won data—to improve the speed and quality of subsequent tasks. This personalization is slated for Enterprise and education users initially.

These substantial upgrades arrive as OpenAI works to solidify its position against rivals like Anthropic, whose Claude Code has gained considerable traction for automating coding work. The latest Codex release demonstrates a focused effort to match and extend the functionalities that are reshaping how developers work.

Source: The Verge