OpenAI's Codex update adds computer use, memory

This update arrives one year after Codex launched, and the article frames it as a natural evolution: developers began using Codex to write code, then expanded into understanding systems, reviewing work, debugging, and coordinating with teammates. The new features — computer use, in-app browsing, image generation, and 90-plus plugins — appear designed to keep Codex relevant across the entire software development lifecycle, not just the coding step.

The release also signals a push toward autonomous, long-running agents. Automations can now reuse conversation threads and schedule work across days or weeks, while memory previews let Codex recall preferences and corrections. These capabilities could reduce the need for developers to repeatedly re-explain context, but they also put more trust in the agent to act independently on a user's machine.

The stakes hinge on reliability: if Codex can handle multi-step tasks without close supervision, it may shift how much groundwork developers delegate to an AI. The phased rollout — computer use starting on macOS only, with EU and UK personalization features pending — suggests OpenAI is managing risk as the agent gains more control over a user's computer.