In this chapter, we examine concrete applications that illustrate how LLMs interact with wireless and mobile network systems. The case studies show how parameter sharing can support efficient model caching at the edge, how multimodal and language-derived environmental semantics enhance service-level traffic prediction, and how post- deployment interaction traces can be distilled into structured experience that improves inference without retraining. We also examine joint optimization of prompt compression and wireless power control for mobile LLM services, the integration of network context into tool routing decisions, and expert-level caching strategies that reduce latency for Mixture-of-Experts models. Together, these applications demonstrate that future ubiquitous intelligence depends on coordinated design across models, memory, and communication resources, enabling LLM-driven services to operate efficiently under real-world network constraints.