This paper is about some difficult issues in systems of systems engineering, especially when the systems are self-modeling. The questions are about which of the available participant systems have the collection of capabilities desired, how to use them to provide those capabilities, and how to avoid introducing new error modes that emerge from the interaction. In this paper, we describe the process of designing and constructing a system of systems, starting with a provided mapping from the perceived need to a coordinated set of system capabilities that can address that need, some of which are missing capabilities for which new systems are needed. The system engineering aspects of this problem are difficult: the available information about the component system capabilities is always incomplete and often inconsistent, and there can be only approximations of the range of environmental behaviors. For these applications, the design will have to change as the environmental conditions change. These systems are in "continual dynamic development", which is why we think the systems should manage and perform the integration tasks themselves. This paper shows one way to do that.