OpenAI gives Codex full control of your Mac

AI coding assistants are great until they hijack your mouse and force you to sit there watching a virtual cursor crawl across your screen.

OpenAI just rolled out a massive revamp to its Codex desktop app, currently used by more than 3 million developers each week. The headline feature is simple but crucial: Codex can now operate in the background on macOS without getting in your way.

Background agents and browser-based markups

Instead of taking over your active desktop, Codex now uses independent background cursors on Mac. Multiple agents can click, type, and navigate desktop apps in parallel while you keep typing in your own windows.

OpenAI pitches this for testing apps, tweaking frontends, or manipulating software that lacks public APIs. It basically turns Codex into a junior dev grinding through repetitive UI checks while you work.

There is also a brand-new in-app browser designed for web and game development. You can comment directly on page elements running on localhost, and the agent reads your visual notes to update the code.

On top of that, OpenAI integrated its gpt-image-1.5 model. Developers can generate mockup visuals, concept art, and placeholder assets directly inside the coding workspace without switching apps.

Memory, scheduling, and over 90 plugins

The update also pushes Codex deeper into company workflows. It now handles GitHub review comments, multiple terminal tabs, and SSH connections to remote devboxes in alpha.

To plug into existing stacks, OpenAI released more than 90 new plugins, with TechCrunch reporting 111 integrations in total. The lineup includes Atlassian Rovo for Jira, GitLab Issues, CircleCI, CodeRabbit, and the Microsoft Suite.

Codex is also gaining a preview of long-term memory to retain coding preferences, alongside scheduled automations that can wake up days or weeks later to finish tasks.

It can even scan unread mentions across Slack, Notion, and Google Docs to hand you a prioritized morning to-do list.

Why it matters

Anthropic has been stealing OpenAI's lunch in the developer space lately, with Claude Code becoming the default darling for engineering teams. This Codex update is OpenAI's direct counterpunch, targeting the exact pain points of desktop automation.

Running agents in the background is a much smarter approach than full screen takeovers. It allows engineers to offload flaky UI testing and asset generation without sacrificing their primary workstation.

That said, we think OpenAI faces a delicate balancing act here. By adding everything from slide previews to Slack triage, Codex is inching dangerously close to enterprise bloatware. If the core code generation cannot match Claude on complex reasoning, all these shiny workflow integrations will just be high-friction distractions.