I knew a guy who could go to the board and draw a FSM, construct a context-free grammar, or construct a proof for anything the professor asked for.

And in Algorithms, he could go to the board and walk through any algorithm we happened to be talking about.

In essence he was perfectly suited to white board interviews.

But when I worked with him in Software Engineering his code was unmaintainable, poorly commented, and he was very abrasive and unpleasant to work with.

Yup. The difference between a computer scientist and a software engineer.

It's a sad state of affairs that our industry is currently conflating the two. Or maybe the industry has only now reached a stage where the two need to become separate?

I mean, you wouldn't expect a theoretical physicist to design an experiment or, worse, a bridge.

I absolutely would expect a theoretical physicist to at least be able to competently contribute to the design of an experiment, especially as it pertains to his specialty.

The bridge analogy is spot on, in my opinion. It mirrors my above comment, and I couldn't agree more.

The theoreticians usually figure out what exactly to look at. The experimentalists figure out how to do it, and then carry it out.

A common joke is that you don't let the theorists anywhere near the apparatus, because they'll break things.

That person knows how to code, he just writes bad code. As most people do at the start of their career. It is learnable though.

The bigger problem apparently is that he's an asshole. Interviews can screen for assholes, but not with programming exersizes.

This perfectly puts an handle on what is happening: http://sijinjoseph.com/programmer-competency-matrix/