Advanced manufacturing control still remains an important topic in current research. Especially aspects of dynamics and of failures in the production process are insufficiently taken into account by systems in use. This paper presents a multi-agent approach to scheduling material flow that shows dynamic and adaptive behaviour. Even though machine scheduling has found a thorough treatment in AI literature, there are only few investigations on the material flow problem. In this paper, it is argued that a decentralized architecture with centralized control fits well with the local and global aspects of the scheduling problem. The top-level algorithms of the scheduling process are outlined and further improvements required are sketched out.