Despite their overwhelming success, present-day Massive Open Online Courses are far removed from the student modelling capacities displayed by earlier Intelligent Tutoring Systems. Being mere content delivery tools, MOOCs typically lack a thorough assessment module as well as tools for personalising the learner's track. When learning music, particularly, these two properties are indispensable. This chapter surveys suggestions made by experts in the field of AI in education today towards the incorporation of ITS tools and techniques into MOOCs. Yet, more traditional student models and tutoring modules are not without shortcomings themselves and the real challenge lies in making active models of both the tutor and the student, which can be used to predict future learning tracks and set the right challenges. Agent-based tutoring systems offer an attractive framework for building such active tutor/student models. The proposed concepts are illustrated in the domain of music composition. A tutoring system has been implemented to teach students the craft of counterpoint, a commonly used strategy for learning polyphonic music composition. It is based on the theory of flow to keep students motivated and optimize learning.