In practice, it often just speeds up the wrong behavior. The app reacts faster, but to the wrong signals.

Have others seen cases where adding AI made the product feel worse, even if the metrics looked okay at first?