Discussions surrounding Industry 5.0 emphasize the need for a fully interconnected industrial ecosystem, integrating AI and digital twins. In this environment, industrial equipment must collaborate seamlessly with human workers, requiring low-latency, high-data-rate connectivity for real-time monitoring. To address this demand, Time-Sensitive Networking (TSN) standards have been developed. However, configuring TSN in dynamic industrial networks poses challenges. The IEEE 802.1Q standard offers mechanisms like the Time-Aware Shaper (TAS) to achieve deterministic latency when configured correctly. In this paper, we tackle the configuration of TAS in dynamic networks, like reconfiguration of production lines to fit production objectives or the deployment of new applications in the production line resulting in adding new flows in the network. Our solution employs Deep Reinforcement Learning (DRL), trained and evaluated through simulations, offering adaptability to changing network conditions and dynamic production line reconfigurations.