In this paper, we highlight what is missing in the approach to architecting and developing AI models that would mean performance is translated into effective hybrid systems. We conclude people, place and purpose should drive new architectures that support rich interaction through tractable representations that will underpin success. We call for the data-driven ML community to embrace the consideration of tractable representations in the architecture of algorithms and place a responsibility on HCI researchers to unwrap and expose the significant factors in the design space that are critical for successful hybrid decision-making in the real world.