The "classical" example is surely duality of abelian varieties. If you want this duality to work over finite fields (or in characteristic p generally), it becomes apparent that you can't work with varieties alone. Technically taking the quotient by a group scheme that is not reduced is just too hard to express in the older geometric language.