This chapter develops a powerful microscopic framework via agent-based modeling for the simulation and prediction of emergent complexity resulting from the dynamic behaviors of and interactions between individuals through digital twins of cities. Using the predator–prey model, it examines urban–rural population dynamics and the evolution of urban structures involving producers and consumers from the deterministic and stochastic perspectives. Cities are considered as multiagent systems consisting of many interacting autonomous agents, each functions with a local view of the environment and self-directed interaction protocols. Because of the high complexity of the adaptive networks with communication topology, it is in general very time consuming and computationally expensive to simulate their dynamics. The chapter shows how to derive reduced-order models which are much lower in dimension yet preserves the structure, stability, consensus or other characteristics of interest in the original large-scale systems. To capitalize on the power of generative AI, instead of crafting a priori rules for communication, we may employ large or small language models to build generative agents who can interact in natural language with one another and the environment and evolve over time through observation, reflection, knowledge updating and planning, resulting in emergent structures.