Human design of any novel system or artifact, e.g. a new type of vehicle, house, or satellite, rests on both general real-world engineering knowledge and inventiveness. At the outset of the conceptual stage of any such design process, requirements are often vague and conflicting—even missing. Consequently, the conceptual engineering design process has proven difficult to implement in machines. Thus, prior work on design automation has not surprisingly focused largely on the later steps of design, which tend to be much more structured. We envision a future of “synthetic engineers” that can help human experts with increasingly complex and challenging systems design. Naturally, we may ask what contemporary research on artificial general intelligence could contribute to this vision. Conversely, we can investigate what goes on in the conceptual engineering design process and ask whether this may provide valuable insights into research on general machine intelligence. Taking this perspective, and based on a review of two existing implemented cognitive architectures, we present a set of what we consider necessary cognitive faculties that must be coherently unified in a single agent architecture to automate the conceptual design process, and a set of minimum requirements that any agent capable of conceptual design must meet.