As cities increasingly rely on AI for sustainability challenges, a critical gap emerges: AI applications in urban planning and safety predominantly proceed without explicit guidance from established urban theories that have guided sustainable development for decades. Our analysis reveals that technology-driven research dominates the field, while problem-driven approaches addressing genuine urban needs remain minimal. To bridge this theory-practice disconnect, we develop a Large Language Model (LLM)-based multi-agent recommendation system publicly available online that realigns AI development with sustainable city principles. The system employs specialized agents to recommend appropriate theoretical frameworks, AI methods, and data sources for urban challenges, drawing from classical urban theories. Through diverse case studies, we demonstrate how our approach transforms technology-focused solutions into theory-grounded interventions that address sustainability’s interconnected dimensions. Our framework fundamentally shifts the question from “what can algorithms do?” to “what does this urban challenge require for sustainable outcomes?”—ensuring AI amplifies rather than replaces the theoretical wisdom essential for creating resilient, equitable, and livable cities that contribute to global sustainability targets.
Tong et al. (Mon,) studied this question.