We expect interactive help in a CLI, and tab completion. We expect errors that tell us what we've done wrong. These things quickly expose the underlying "shap

We expect interactive help in a CLI, and tab completion. We expect errors that tell us what we've done wrong. These things quickly expose the underlying "shape" of the interface. When chatting with an AI, you don't know that the bot was 41% sure that this was the right path for you, chosen out of five other options with lower scores. It just takes you where it thinks you want to go without sharing that structural information with you.