Product-driven manufacturing has gained a lot of traction recently among practitioners as it has the potential to take flexibility and agility of the manufacturing system to a new level compared to hierarchical control models. The advances in embedded technology have created the premises for the emergence of truly intelligent products that are capable not only of identification and information storage, but also of complex behavior and local decision making. In this context, this paper proposes a multi-agent control system that aims to solve the new challenges introduced by the shift to product-driven manufacturing, specifically addressing the special needs for information flow between shop floor entities and the MES system. The paper presents the pilot implementation, using the JADE multi-agent platform, a backtracking scheduler, an artificial neural network (ANN) for local decision making and the experimental results outlining the agent processing requirements during the product lifecycle.