They measure something different, I agree, but not different enough to matter in real life decision-making process, which often involves factors outside of the model.

In real life decision-making, there are many other factors that are not known or quantifiable that come in and dominate any errors arising from using confidence instead of predictive interval.