Quality attributes in non-AI systems are in what I would call the “adolescent” stage. That is, we have definitions and architecture tactics for improve the performance of individual quality attributes. What we do not have is a unified theory of quality attributes. Each QA is based on different considerations. A unified theory would put performance, security, verifiability, availability, and other QAs on a footing that allows modeling them all in the same terms and recognizing trade offs in an analytic fashion. In AI systems, QAs depend not only on the software architectur but on the data used to generate and train the models. What are the techniques that will enable the construction of a list of tactics for QAs involving data? In this working session we will explore which quality attributes need immediate attention from the software architecture community to further detail with its corresponding tactics and patterns to provide better guidance to development of AI-enabled systems.