OpenAI Adds Record & Replay to Codex for macOS Business Users

The most interesting thing about OpenAI's Record & Replay for Codex is not what it records, but who it's for. Most AI coding tools assume you can describe what you want precisely enough for a model to act on it. Record & Replay flips that assumption: if the workflow is easier to show than to describe, you do it once on your Mac and let Codex learn by watching.

According to OpenAI's ChatGPT Business release notes, the feature lets eligible Business users demonstrate a workflow once and turn it into a reusable skill. During recording, Codex observes the actions and window content needed to learn the workflow, and the captured skill can then be replayed with Codex, Computer Use, browser actions, plugins, or a combination of available tools, no re-recording required. The practical examples given include filing an expense, booking a parking space, creating a correctly configured issue, publishing a video, or downloading a recurring report. Those are not developer tasks. They are the unglamorous operational workflows that run every business but rarely get automated because scripting them is too much work relative to their frequency.

This matters because it pushes Codex into territory traditionally occupied by robotic process automation tools, a category that has long owned the "record your screen, replay it forever" market for enterprise automation. The distinction Codex brings is the ability to blend a recorded workflow with reasoning, handing off to browser actions, Computer Use, or plugins as the task requires, rather than executing a brittle pixel-for-pixel playback. Whether that makes replayed skills genuinely more reliable, or just fail in more interesting ways, remains to be seen.

The honest caveats are worth sitting with. Record & Replay requires Computer Use to be available and enabled, which narrows the eligible audience at the outset. Geographic availability at launch excludes the European Union, UK, and Switzerland, meaning any organization spanning those regions cannot roll it out uniformly. OpenAI's own guidance to keep recordings focused on the workflow and avoid secrets or sensitive data is the kind of warning that suggests accidentally capturing credentials or PII in a screen recording is a real risk, not a theoretical one. What the release notes do not give you is how these recorded skills are stored, who can access them at the org level, and what happens when a third-party UI changes and the recorded workflow no longer maps to reality.

For teams that have shelved automation projects because building reliable scripts was too expensive relative to the payoff, this is worth a close look. The expansion from code generation into repeatable cross-tool workflow execution is a meaningful shift in what Codex actually is, and it lowers the floor for non-technical operators to build real automation without touching a prompt template.