One paper accepted to EMNLP! We build an interface to help users solve complex questions with plans and show predicting which plans help humans is difficult for humans, reward models, and agents!