OpenAI’s big Codex update is a direct shot at Claude Code
Codex can now autonomously operate macOS apps in the background while multiple agents work in parallel.
OpenAI is escalating its rivalry with Anthropic by releasing a significant update to its Codex agentic coding system. The centerpiece is a new capability allowing Codex to autonomously operate desktop applications on a user's macOS computer. This background operation enables developers to test frontend changes or work with apps lacking APIs without interrupting their primary workflow, and supports multiple parallel agents. The feature is rolling out first to ChatGPT desktop app users on macOS, with EU availability and expansion to other operating systems promised for later.
Beyond app control, the update equips Codex with several powerful new tools. It gains image generation and iteration abilities through integration with gpt-image-1.5, and adds native web browsing with an in-app browser for direct page commenting. New plug-ins connect Codex to developer staples like GitLab, Atlassian Rovo, and the Microsoft Suite. Crucially, Codex now includes a memory feature (in preview) to retain user preferences and corrections, and can schedule future work to automatically resume long-term tasks. These personalization and automation upgrades aim to reduce the need for detailed custom instructions and speed up task completion.
- Codex gains autonomous control of macOS desktop apps, operating in the background with support for multiple parallel agents.
- New features include image generation via gpt-image-1.5, plug-ins for GitLab/Microsoft Suite, and native in-app web browsing.
- A memory system (in preview) allows personalization, and Codex can now schedule and automatically resume long-term tasks.
Why It Matters
This transforms Codex into a proactive development assistant that can automate complex, multi-step software tasks and UI testing directly on a user's machine.