Recall that in a top$-$down insertion into a Red$-$Black tree, the rotations move nodes off of the path of insertion, which allows us to eventually add a node to the path of insertion without increasing the height of the tree. Your friend suggests that you instead do away with colored nodes and always rotate the tree away from the path of insertion as you move down the tree. Why would this not be desirable?

For deletions, we would need to perform analogous rotations away from the path of deletion, leading to a high constant factor for these balanced trees.

Unlike Red$-$Black trees, which only perform rotations when necesesary to re$-$balance the tree, this new scheme can perform extraneous rotations.

If the path of insertion is long, we will rotate too many nodes off of the path$-$of$-$insertion, leading to an imbalanced tree.

As we traverse down the tree, we cannot know precisely where the point of insertion will be$.$