Not a surprise, really.

I have battled so many times with smug novice "engineers", fresh from their CS courses, pointing, without even thinking twice, O(n^2) or even O(n^3) algorithms as errors on code reviews. This without taking into account the size of the input or the underlying machine executing the code.

It is my constant source of amazement -- to interview yet another candidate who can write any number of sorting algorithms from memory on a whiteboard but who can't tell whether a linked list or an array list insertion is going to be faster for a given application.