The paper aims to provide high-level guidance for architects of cyber-physical enterprises. We propose that interactions within such systems should be largely self-determined, based on system self-awareness and dynamic re-configuration, with the architecture evolving based on a set of foundational principles, rather than being pre-defined by an external designer. We investigate the suitability of typical development life cycles and identify architectural challenges in the context of dynamic cyber-physical systems that utilize the power of the Internet of Things. Desired systemic attributes are defined, which are necessary for making suitable core architectural choices. The application of the findings is exemplified through a case study, a synthesis of issues, and implications for further research.