The fast growth of smart cities has led to more traffic congestion, made worse by IoT devices providing continuous high volumes of information. Good traffic control enhances the mobility of cities and also brings down environmental pollution and supports energy saving. The research presents a new design that uses artificial intelligence (AI) agents along with large language models (LLMs) to organize high-level strategies and daily manage the operation of cities’ traffic networks. The system is based on advanced learning methods, flexible scheduling systems and predictions for route planning to save energy and prevent traffic jams when road congestion is high. The main part of the architecture consists of macro agents that control and coordinate with small local agents throughout the city. The system depends on vehicular sensors, GPS systems for location tracking, environmental devices and simulators for VANETs. The hierarchy of LLM agents in traffic control monitors traffic, develops clear rules for managing it and decides probabilistically to regulate traffic dynamically. Real-time data exchange between agents is made easy by a strong mesh network which also keeps delays and mishaps low. On embedded urban traffic information, experiments using simulation test how resources are being managed by tracking energy consumption per node, idle time for vehicles and the efficiency of transport activity. This architecture is proven to lower congestion and energy usage, demonstrating it can be scaled up and deployed in real time to support green and efficient travel in future cities.