Several information technology (IT) artifacts exist that can benefit sustainable development, however were not developed under adequate consideration of design science research methodology. In response, we present a methodological contribution by introducing a qualitative reflective theorizing approach that derives design principles from already existing IT artifacts with grounded theory procedures. We validate our approach by investigating the special case of communicating properties of artificial intelligence (AI) models via high-level consumer labels, drawing inspiration from well-established systems such as energy labels and Nutri-Score. Such AI energy labels inform diverse stakeholders on the predictive quality and resource consumption of models in comprehensible ways, however design knowledge has not been properly captured for their development. Building on existing reflective practices like design archaeology, we conceptualize how adequate design principles can be developed based on existing artifacts in-situ, grounded in results from a qualitative interview study. Our research offers insights into reflective design processes and proposes four actionable design principles for advancing transparency and sustainability in AI.