Maintenance has become critically important for industrial sectors due to the increasing complexity of interactions among various production activities. The shift from predictive maintenance to prescriptive maintenance (PsM) marks a significant evolution, as PsM goes beyond forecasting to provide specific, actionable recommendations. It prescribes what actions should be taken, when they should be performed, and how they should be executed, leveraging advanced analytics and machine learning (ML) techniques. Prescriptive maintenance is considered the most advanced and intelligent form of maintenance, offering a holistic approach that integrates planning, decision-making, and adaptive learning within the context of smart manufacturing. This transition introduces a new perspective on flexibility, customization, and resilience in production planning. These capabilities are made possible through the application of advanced data analytics and ML models, which are capable of processing vast amounts of data from both IoT (Internet of Things) devices and historical records, thereby enhancing the precision and reliability of proposed maintenance interventions. To address these objectives, this work proposes a decision-making framework whose core foundation integrates agentic artificial intelligence (Agentic AI) with the digital twin paradigm. Compared to alternative approaches, the integration of Digital Twin technology and Agentic AI represents a notable advancement in the field of prescriptive maintenance systems. This integration enables the creation of virtual testing environments where maintenance strategies can be simulated, evaluated, and optimized prior to their implementation in real-world industrial contexts.