Numerical models contribute to a substantial part of research conducted in Earth sciences. To be executed on a computer, the mathematical representations they depict are transformed into a computer program. The present article aims at following this transformation and at questioning the implications of rendering a representation computationally executable. It uses to this end a case study of the so-called “sticky air” numerical method employed in geodynamics, which consists of putting a layer of a hybrid material — a hundred thousand quintillion (1023) times more viscous than air — on top of the Earth crust. Far from being only a marginal and automatic step in the modelling process, the constitution of an executable computer code is the result of a profoundly creative and negotiated practice.