Smart cities increasingly rely on heterogeneous and interdependent digital infrastructures combining mobility, energy, environmental monitoring, water and waste management, cybersecurity, telecommunications, public services, and critical infrastructure supervision. Although specialized artificial intelligence components can provide high performance within individual domains, city-scale intelligence raises a broader architectural problem: how to determine which capabilities should be activated, how to coordinate them across domains, how to prevent persistent activation of unnecessary computational resources, and how to preserve an interpretable memory of the context that produced each operational response. This article introduces S-AI-CITY, a domain-specific instantiation of the Sparse Artificial Intelligence (S-AI) paradigm for adaptive urban systems. The proposed architecture is based on four complementary principles: activation parsimony, artificial hormonal orchestration, functionally specialized urban agents, and persistent symbolic memory. Six canonical urban hormones are introduced: Mobilin, encoding mobility and transportation stress; Energexin, representing energy and resource pressure; Environin, representing environmental and utility stress; Resiliencin, encoding infrastructure failure, emergency, and recovery pressure; Normin, representing cybersecurity, privacy, trust, and compliance stress; and Servicin, representing public-service demand, collective-event pressure, accessibility, and citizen-facing service load. Urban Gland Agents transform normalized observations into bounded hormonal emissions. A Hormonal Engine integrates these emissions, applies decay, crossregulation, and topology-dependent propagation, and produces the shared hormonal state 𝐻(𝑡). The CityMetaAgent then uses this state