Under which conditions does simulation-based learning facilitate the development of complex skills in higher education? Through which learning processes and for whom is simulation-based learning effective? In this lecture, I will present a conceptual framework, linking learner characteristics, learning processes, and learning outcomes, while considering instructional support designed to facilitate learning processes in simulations. Building on this framework, I will present methods and findings from experimental and meta-analytic studies conducted within the DFG Research Unit FOR 2385 COSIMA. Findings address prerequisites such as prior knowledge, effective learning processes in simulations, including repeated engagement in epistemic activities, and the effects of instructional support through targeted scaffolding and feedback. Differences across simulation types, domains, and levels of prior knowledge will also be explored.  A particular focus will be on how artificial intelligence can support both data analysis and the provision of personalised instruction, such as adaptive and adaptable scaffolding and feedback within simulation-based learning. Finally, I will discuss the extent to which the conceptual framework is supported by the evidence, remaining open questions, and possible trajectories for future cross-disciplinary research on AI-supported simulations in higher education.