A new class of application that operates independently of direct human control is starting to emerge. It is our belief that the development of such applications is highlighting the shortcomings of current communication architectures and middleware infrastructures. In particular, they do not adequately support advanced dynamic interaction models, e.g., in the field of autonomous agents, distributed AI, and mobile co-operating entities. As we describe, our work represents the beginning of an attempt to bridge the gap between the requirements being put on system support by these advances, and the shortcomings of current architectures and middleware models. Therefore, this paper explores the problem of providing infrastructure support for large distributed systems composed of mobile autonomous components. It describes a programming model supporting these applications based on the concept of sentient objects, and a hierarchical distributed communication architecture. Because the components may be part of the physical environment, issues such as predictability and environment awareness in the interactions between objects and environment deserve particular attention.