The purpose of this paper is to determine a set of requirements and proposals for the design of a collaborative interactive system working under distributed tangible environments. A wide range of human-machine, human-human and human-environment interactions is to be considered, and using multi-agent architectures was recommended since they have already shown their benefits in developing large and complex systems under a unified framework. We further argue that such a design would require an in-depth analysis of four main properties (autonomy, proactiveness, context-awareness, situatedness). In this context, the potential interest and complementarity of two main design approaches (holonic and normative) are discussed. To support this discussion, two applications are also described – a Road Traffic simulation and a version of the game of Risk. Both of these applications introduce complementary man-machine-environment interaction issues