The idea is to run a slow, expensive computer use agent once and record each of the steps it took. Then we can rerun the cached actions again and again.
Could be useful for QA testing or RPA maybe.
I'd love to hear people's feedback or ideas!