Their interview process itself makes very little sense though. There's literally nothing in it that resembles the job you'll be doing on a day to day basis (copying from one protocol buffer to another, while sipping a free latte), so they're hiring for some proxy traits the correlation of which with job performance is low to non-existent. They know this, too. They've done a study where they just hired some number of people at random, without looking at their interview results, and then tracked their performance over the next few years. Those folks performed about as well as people who received high interview scores.