Networked smart objects will be called on to perform multiple sophisticated tasks, and thus must manage new levels of complexity while minimizing energy use. This paper argues that today's dominant architectural vision of networked things, which is closely related to wireless sensor node architectures of years past, should be re-examined in light of key functional requirements: energy efficiency, real-time processing, and composability, as well as heterogeneity and evolvability. The traditional approach is to add software complexity to achieve all but the first, but the associated overhead incurs an energy cost, degrading device lifetime. An alternative approach is to harness, at the system level, recent and significant improvements in low-power technologies. Here we demonstrate a re-thinking of smart object design based on the goal of minimum life-cycle cost, and show how it motivates the organizational principles of specialization and division of labor. These coupled principles in turn suggest a functional architecture for smart objects that integrates two system-level properties: parallel processing and hierarchical power gating. We describe a prototype architecture and implementation that incorporates this design approach, and discuss its benefits to smart object design.