Another way to put it: If you know how long something will take in advance, you have a solution in mind. It is unlikely that this solution is (A) the best one and (B) the one you will actually implement. It would be stupid to ignore information you learned along the way. If you could actually predict the future you should invest in the lottery, not in software.

EDIT: Of course there are projects where you actually know exactly what to do. Happens a lot in consulting. That has nothing to do with Agile though.