This part remains me too much of an interview: somebody sitting next to you and waits in the real time when you say something. It is too much pressure for me, somehow.
There isn't much benefit to pair programming. It does catch errors sooner. It does help structure code better. It does eliminate typos the compiler can't catch. So it does have benefit. But I think the cost of two programmers does not justify the benefit.
The real problem is 'shared credit'. Remember those school projects where 2 or 3 people worked together? I had one in grad school where I ended up writing almost the whole project myself but we all got an 'A' grade. Management has no way to figure out how to allocate credit. This is a problem in shops where they put people on a scale of 1-5 or some such nonsense. If your shop does this, do NOT pair program.