I will describe a recipe for predicting certain phenomena in the Langlands program which takes the following form: 1) Start with class field theory. 2) Find its linear-algebraic underpinning. 3) De-linearize it.
