The rapid expansion and complexity of modern network infrastructures, encompassing telecommunications, cloud, and IoT ecosystems, have intensified the need for intelligent, scalable, and adaptive optimization strategies. Multi-agent AI systems, comprising multiple autonomous and collaborative agents, have emerged as a promising approach for addressing these challenges by enabling distributed decision-making, real-time adaptability, and collective intelligence. These systems allow agents to coordinate, negotiate, and learn from local and global network conditions, thereby improving resource allocation, traffic routing, fault management, and overall network performance. By leveraging techniques such as distributed reinforcement learning, swarm intelligence, and game-theoretic coordination, multi-agent AI systems can dynamically optimize network operations while mitigating congestion, latency, and energy inefficiencies. This review explores the conceptual foundations, architectural frameworks, enabling technologies, and optimization strategies of multi-agent AI systems in large-scale network environments. It further examines practical applications, including cloud networks, IoT deployments, and telecommunications, highlighting measurable improvements in efficiency, scalability, and resilience. Finally, the paper discusses key challenges such as communication overhead, coordination conflicts, and security risks, while outlining future research directions that include autonomous network orchestration, edge intelligence integration, and AI-enhanced 6G networks. The findings indicate that multi-agent AI systems are critical for enabling self-optimizing, adaptive, and cost-efficient networks, positioning them as a cornerstone for next-generation network management.