This is true today and has been my experience as well -- both to write software as well as doing computational physics.

The interesting question then is to ask how long will this stay true? As the models get better will they eventually not need the human expertise to start adding value?