Simulating human activities in an urban setting requires a fundamental understanding of what human behaviors are likely, normal, or anomalous in such an environment. While many CG research groups are producing animated crowds of thousands of people, generally the agents are mostly just pedestrians wandering (often aimlessly) the traversable areas. In contrast, our CAROSA system, created by Jan Allbeck, builds on top of computer graphics “crowd” simulations in our HiDAC simulator by adding an action ontology and agent models that together create data-driven, purposeful, functional individual agents. The CAROSA methodology includes scheduled, opportunistic, reactive, and aleatoric (stochastic) actions, defined with common user interface tools, mediated through a common parameterized action and object representation (PAR), and executed by a flexible real-time behavioral engine. To trigger agent behaviors without heavy-duty reasoning or planning, we developed an extension of smart objects called "smart events" that inform agents of tasks that require their attention and participation. Agent responses also depend on their current role or priming to nearby agents and context. Examples of such behaviors include dyadic conversations. Ongoing work includes developing these concepts into narrative structures to give interactive life and meaning to urban settings.