This paper proposes an Agentic AI-driven dual spatial-scale optimization framework for Unmanned Aerial Vehicular Networks (UAVNs) deployment. It leverages exact potential games (EPGs) at different scales and integrates a large language model (LLM) to enhance utility weight generation, leading to improved energy efficiency, latency, and throughput.