OpenAI Launches the Codex app, their take on Agentic Coding
OpenAI released Codex for macOS today, a desktop app for managing AI coding agents. The app is designed as a command center where you can run multiple agents in parallel, each working on separate tasks without conflicts.
What It Does
new text here preview
Codex organizes work by projects in the left sidebar. Each project can contain multiple chat sessions where agents handle different tasks building features, fixing bugs, running refactors. The agents work in isolated worktrees, so their changes don't interfere with your local git state until you review and approve them.
The toolbar includes buttons for opening your development tools, committing changes, accessing a terminal view, and reviewing code diffs. Settings let you customize behavior, connect MCP servers, and integrate skills.
Skills and Automations
Skills are packaged instructions, resources, and scripts that agents can execute reliably. OpenAI is launching with an open-source library of skills in their GitHub repo, covering common workflows like converting Figma designs to code, managing Linear issues, and deploying to Vercel or Cloudflare.
When you create a skill in the app, it syncs across all surfaces where you use Codex like in the desktop app, CLI, and IDE extensions. You can also check skills into your repository to share them with your team.
test
Automations run on schedules you define, handling repetitive tasks in the background. At OpenAI, they're using automations for daily issue triage, summarizing CI failures, generating release briefs, and monitoring for bugs. When an automation completes, results land in a review queue so you can jump back in if needed.
test
Availability and Pricing
The Codex app is available starting today for macOS. You can download it from OpenAI's website. Windows and Linux versions are planned.
Codex is included with ChatGPT Plus ($20/month), Pro, Business, Enterprise, and Edu subscriptions. For a limited time, OpenAI is making it available to ChatGPT Free and Go users, and they're doubling rate limits for all paid plans. These higher limits apply everywhere you use Codex wether it's the app, CLI, IDE extensions, and web interface.
Usage has grown substantially since the launch of GPT-5.2-Codex in mid-December, with overall usage doubling and more than a million developers using Codex in the past month alone.
Key takeaways
The interface is clean and purpose-built for agent supervision rather than traditional code editing. While testing, I had Codex perform a code review of one of my projects. It spent time examining files, then asked permission to run commands outside its sandbox when needed.
The worktree support is particularly useful. You can explore multiple implementation approaches simultaneously without impacting your main branch. The diff view makes it straightforward to review exactly what each agent changed before merging.
OpenAI says Codex is "the most loved internal product we've ever had" at the company. Based on this initial look, the focus on workflow orchestration and background automation makes sense for teams managing larger codebases. It'll take more time to fully assess how the skills and automation features perform in practice, but the foundation is solid.