Limitations of Centralized Smart City Intelligence: A Literature Review toward Federated, Explainable, and Agentic AI

Smart city systems are starting to use centralized intelligence to collect, process, and analyze urban data on a large scale. This occurs so that smart cities can help coordinate and manage across multiple domains, including transportation, energy, and public services. Centralized intelligence provides the ability to monitor integration and provide management coordination. However, this approach creates challenges in terms of privacy concerns, limited scalability, communication overhead, a single point of failure, lack of transparency, and limited adaptability in changing environments. This research work delivers a comprehensive literature review to examine the limitations of using centralized intelligence for smart cities and investigate new ideas on how to address those limitations. This review includes a focused examination of recent advances that have been made toward improving distributed learning, explainable artificial intelligence, and agentic intelligence to support urban decision-making.  The evaluation shows that existing approaches independently address privacy, transparency, and adaptability but lack coordinated integration. This finding shows there is significant research needed to identify a unified, comprehensive framework that includes all of these characteristics. The results of this review and assessment contain the groundwork for future studies toward creating integrated, scalable, transparent, and adaptive intelligent decision-support systems to create sustainable smart cities.