In critical wireless communication scenarios like autonomous driving and industrial automation, the demand for ultra-reliable and low-latency communication (URLLC) is paramount, imposing stringent criteria for reliability, latency, and jitter. Traditional medium access control (MAC) protocols encounter limitations due to their inflexible scheduling mechanisms. This study introduces an innovative adaptive AI-driven scheduling framework for URLLC, harnessing deep reinforcement learning (DRL) to dynamically adjust scheduling in response to evolving conditions. Through extensive training in simulations with critical metrics rewards, the DRL agent strategically optimizes trade-offs between reliability and latency. Results unequivocally demonstrate the framework's superiority over conventional protocols, significantly enhancing reliability, latency, and jitter performance. Its adaptability shines through its ability to accommodate a diverse range of network dynamics and traffic patterns. This pioneering approach effectively addresses the exigencies of mission-critical communication, thereby bolstering wireless technology's capacity for responsiveness, reliability, and adaptability to unprecedented levels.