Military planners would benefit greatly from a planning system that dynamically incorporates real-world domain information, allowing plans to adapt easily to new information. The Joint Air Ground Unified Adaptive Replanning System (JAGUAR) system provides a model-driven planning paradigm for such a dynamic air mission planning domain. Developing models for this distributed and dynamic environment proved challenging, both for building models that were meaningful for all of the components and for creating models that could be fed in seamlessly to the running system. This paper gives an overview of the lessons learned during the six-year effort of developing adaptive models for the JAGUAR system.