Composing Grid Services to workflows, which solve specific problems is not a trivial task. Various parameters like for example timing constraints, desired quality of solution, availability of (computing) resources, error recovery, payment and security issues, to name only a few, have to be considered. The heterogeneity and the highly dynamic nature of the Grid pose additional challenges. Current implementations try to construct workflows based on the user's needs and the capabilities of the respective services before the actual run. In this paper we propose a blackboard method that is able to find (near) optimal workflows and accommodate to changing conditions like for instance delays or faults on the fly.