This paper presents a comparative study of four multi-agent system architectures—centralized, hierarchical, shared message pool, and decentralized—applied to a telecom customer service scenario. Using a consistent set of specialized agents, tools, and knowledge bases, we implement and evaluate each architecture to analyze differences in coordination mechanisms, information sharing, task workflows, extensibility, and robustness. The study demonstrates how architecture choice impacts system behavior, from tightly controlled centralized coordination to fully autonomous decentralized collaboration. By illustrating a practical telecom use case, this work bridges theoretical multi-agent system models with practical enterprise applications, offering insights for designing AI-driven service systems. This work not only provides guidance in selecting multi-agent architecture, but also delivers a telecom customer service application case, demonstrating how multi-agent system concepts can be effectively implemented in practice.