Codex's new Record & Replay feature is getting a lot of attention this week, show it a task live, it converts the demo into a reusable skill. It's a good feature, currently Mac-only and tied to Codex for execution.

I built the same core idea into RAPR AI about a month ago, with one deliberate difference: instead of live screen capture, I use a recorded video as the input, any screen recorder works, nothing proprietary required. You record the task once, hand it to RAPR AI, and it generates a playbook from it.

The thinking behind that choice: live capture means the model is active for the entire demo, including the dead time, the searching for a file, the pause to think. That's tokens spent on noise. A recorded video can be trimmed and reused, which makes the whole process cheaper and the resulting playbook more reliable, since you're feeding it your best take, not your first take.

The other piece is model flexibility. The playbook isn't locked to one AI. You can execute it through Claude, Gemini, or Codex.

If you're automating repeatable workflows and want something that isn't tied to a single vendor or a single OS, this is the kind of thing RAPR AI was built for. Glad to walk through how it works.