The exponential growth of global data traffic has underscored the limitations of traditional routing protocols such as BGP, OSPF, and MPLS, which rely on predefined policies and reactive measures. While BGP dynamically adapts to network changes, its convergence time and path selection mechanisms often lead to suboptimal routing, increased latency, and inefficient bandwidth utilization. As network infrastructures scale, these approaches struggle to adapt dynamically, resulting in suboptimal routing decisions, increased latency, and inefficient bandwidth utilization. This paper introduces AI-driven routing protocols that leverage machine learning (ML) and software-defined networking (SDN) to optimize real-time network traffic. I propose an autonomous routing framework integrating reinforcement learning-based path optimization, predictive congestion control, and self-healing network mechanisms to enhance network performance by up to 200%. My experimental simulations in Mininet and GNS3 demonstrate a 35% reduction in packet loss, 40% improvement in traffic predictability, and sub-second failure recovery. Additionally, I discuss the scalability of AI-enhanced routing in real-world network infrastructures, emphasizing its applicability in 5G/6G networks, cloud environments, and enterprise SDN deployments.