This Note contains the text of an oral presentation delivered to the third "Thinking Red in War Games" workshop held at the National Defense University in June 1988. It describes some of the lessons learned from developing political-level and theater-commander-level decision models for use in the RAND Strategy Assessment System (RSAS). These knowledge-based models (called Red Agents) can represent alternative mindsets and alternative military strategies. The author begins with a short summary of RSAS architecture and a discussion of reasons for having Red Agents. He then illustrates the use of Red Agents with two examples: (1) an assessment of flexible response that illustrates how the rigor required in modeling can change fundamental perceptions about escalation and deescalation, and (2) a representation of alternative theater-level Red operational strategies for force employment in the Central Region that demonstrates how critical the strategy variable is in analysis, and shows that it can now be explored, rather than left buried in the script-like databases that drive most combat simulations. The author concludes that working with Red Agent models is valuable, and has enormous, even revolutionary, potential.