We have been modeling an ever-increasing scale of applications with agents that simulate the pattern of life (PoL) and real-world human behaviors in diverse regions of the world. The goal is to support sociocultural training and analysis. To measure progress, we propose the definition of a measure of goodness for such simulated agents, and review the issues and challenges associated with first-generation (1G) agents. Then we present a second generation (2G) agent hybrid approach that seeks to improve realism in terms of emergent daily activities, social awareness, and micro-decision making in simulations. We offer a PoL case study with a mix of 1G and 2G approaches that was able to replace the pucksters and avatar operators needed in large-scale immersion exercises. We conclude by observing that a 1G PoL simulation might still be best where large-scale, pre-scripted training scenarios will suffice, while the 2G approach will be important for analysis or if it is vital to learn about adaptive opponents or unexpected or emergent effects of actions. Lessons are shared about ways to blend 1G and 2G approaches to get the best of each.