His idea of pair programming as interview is actually pretty good. However I would stress that there is no such thing as bad interviews, just bad interviewers. I've seen cases of "one secret right answer", "question with hidden land mines" and "irrelevant puzzles that you would never need to solve on actual job". My favorite strategy is to think of some problem I solved myself as part of the job, abstract it out a bit and see how candidate would go at it. This makes things very relevant, allows me to compare candidate's thinking and performance with my own and avoids asking popular chewed up practice puzzles. The problem is that this puts a burden on interviewer to first think of great problem they had solved and then nicely make it simple, interesting and friendly question. This is much harder than looking up one of the irrelevant popular puzzles asking one of them.