OpenAI lets Codex turn screen recordings into automated skills

Teaching an AI to handle mundane office chores usually requires fragile scripts or endless prompt tweaking. OpenAI now wants you to just show it what to do.

OpenAI rolled out Record & Replay for its Codex platform, a new feature designed to turn manual workplace routines into reusable automation.

Show it once, run it forever

The mechanic is straightforward: you hit record and walk through a repetitive task on your screen, such as filing an expense report or submitting a time-off request.

You retain complete control over when recording starts and stops. Once you finish, Codex converts that live demonstration into an inspectable, editable skill.

Teams can then collect these skills into a shared library, letting colleagues rerun identical workflows without repeating the manual legwork.

Why it matters

We see this as a clever push directly onto traditional RPA turf. By grounding the automation in real user demonstrations, OpenAI bypasses the hassle of writing brittle bots or debugging vague natural language prompts.

Crucially, the generated skills are explicitly inspectable and editable. It does not treat workplace tasks like an opaque black box; if a form field changes, users can actually step in and adjust the instructions. For everyday knowledge workers, this drastically lowers the friction of turning boring, repetitive clicks into reliable company-wide tools.