This paper introduces the application of an unconventional approach to Grid programming. The proposed programming model is based on the chemical metaphor for expressing coordination of large grain computations. A chemical program can be seen as a set of chemical reactions, representing computations, that transform a set of floating molecules, representing data, within a chemical solution until an inert solution is reached. This model is intrinsically parallel and possesses nice autonomic properties, which are expected for programming Grids. We illustrate this novel programming model with a simple ray-tracing application and describe an implementation in the context of Desktop Grids.