The first time I met Rob, he was spouting some utter nonsensical drivel. So I did the proper thing and told him that what he suggested wasn't possible, and if he thought it was, he could go and code it.
That ended the conversation as far as I was concerned, and I was quite sure that I'd never hear from him again. Less than 24 hours, he was back, and he had a solution coded, for the problem I considered impossible. After I picked my jaw off the floor, I was quite impressed.
That feature is now one of the major distinguishing feature for RavenDB (dynamic queries). Since then, I have had the pleasure of working with Rob on several occasions. It has been a pleasure, because there is zero nonsense to deal with, just a lot of good code, flying by very quickly.
The last time I had a chance to work with Rob, he stayed for a week, and we worked for six months to catch up. Large parts of the RavenDB 2.5 release was written by Rob during that single week.
He is one of the people in the short list I would go to if I needed help or feedback, from "how do I design this distributed system to be reliable & fast" to "can you help me trace this race condition?".