On one of my previous jobs one problem domain was some area of theoretical physics. There was a guy with brilliant understanding of the problem domain. He had couple of decades of experience teaching physics in the country’s best university. He could explain very hard problems in easy to understand way. The company filed multiple patents with the amazing stuff he invented.

It didn’t help the guy to write code. It wasn’t good. Not only unmaintainable, also slow and unstable (not all algorithms work equally well when mathematically-real numbers are substituted by software-real IEEE754 floats). We ended up just throwing out whatever he coded, and instead learned enough problem domain from him to be able to understand and implement what he actually was trying to do.