Intelligent environments frequently embed a varying number of output means. In this paper, we present an analysis of the task of generating a coherent presentation across multiple displays. We review problems and requirements, and propose a multi-layer approach aimed at addressing some of the previously identified issues. We introduce a low-level architecture for the handling of devices as well as mechanism for distributing coherent presentations to multiple displays. We then discuss what further steps are needed on the way from an abstract presentation goal to a concrete realization.