OpenAI Codex: Mac Control and Image Generation Apfelpatient

OpenAI has released a comprehensive update to its AI coding agent, Codex. The Mac app can now independently control desktop programs, remembers recurring workflows, and generates images directly within the app.

The competition among AI coding assistants on the Mac is intensifying. Just yesterday, Anthropic released its Claude Code redesign with parallel sessions and routines, along with the new Claude Opus 4.7. Now, OpenAI is countering with a major Codex update that significantly expands the functionality – and includes a feature directly reminiscent of Anthropic's computer-use approach.

The most striking new feature: Codex can now autonomously control desktop apps on the Mac. The agent sees the screen content, moves the cursor, clicks, and types—all without user intervention. Multiple Codex agents can run simultaneously without interrupting the user's work. OpenAI particularly recommends this feature for testing apps, iterating frontend changes, and automating repetitive tasks.

Memory and long-term automation

Codex now remembers user preferences, recurring workflows, tech stacks, and other project-related information. This means that the longer you work with Codex, the less context you need to provide for new tasks.

Even more interesting is the new long-term automation: Codex can resume work after a break via existing conversation threads, schedule future tasks itself, and work on a task for days or weeks. It also independently suggests new tasks based on project context, memory, and connected plugins.

This is a significant step beyond what Claude Code's new routines offer – automations still need to be configured manually there.

In-app browser and image generation

A new in-app browser allows users to leave comments directly on websites, giving the agent more precise instructions. In the future, Codex will be able to fully utilize the browser – opening websites, simulating user flows, taking screenshots, and inspecting outputs.

Codex now also uses OpenAI's gpt-image-1.5 image generation model to create images directly within the app. This is especially useful for developers who need placeholder graphics, icons, or visual prototypes for their projects without having to switch apps.

Claude Code vs. Codex: The race intensifies

Anthropic and OpenAI are neck and neck in the race to become the best AI development environment on the Mac. Both platforms offer parallel sessions, computer-use capabilities, and project-spanning memory. Codex focuses more on standalone long-term automation, while Claude Code scores points with its revamped interface and integrated terminal.

For Mac users looking to integrate AI into their development workflow, this means both tools are improving rapidly – and there's never been a better time to get started. (Image: OpenAI)