Steel production is a complex process and finding coherent and effective schedules for the wide variety of production steps, in a dynamic environment, is a challenging task. In this paper we propose a multi-agent architecture for integrated dynamic scheduling of the hot strip mill and the continuous caster. The scheduling systems of these processes have very different objectives and constraints, and operate in an environment where there is a substantial quantity of real-time information concerning production failures and customer requests. Each process is assigned to an agent which independently, seeks an optimal dynamic schedule at a local level taking into account local objectives, real-time information and information received from other agents. Each agent can react to real-time events in order to fix any problems that occur. We focus here, particularly, on the hot strip mill agent which uses a tabu search heuristic to create good predictive-reactive schedules quickly. The other agents simulate the production of the coil orders and the real-time events, which occur during the scheduling process. When real-time events occur on the hot strip mill, the hot strip mill agent might decide whether to repair the current schedule or reschedule from scratch. To address this problem, a range of schedule repair and complete rescheduling strategies are investigated and their performance is assessed with respect to measures of utility, stability and robustness, using an experimental simulation framework.