In a Problem Solving Environment (PSE), a scientific workflow management system (SWMS) provides a meta environment for managing activities and data in scientific experiments, for prototyping experimental computing systems and for orchestrating the runtime system behaviour. A Grid infrastructure makes data and computing intensive experiments feasible in PSEs but also requires the management of workflow to support dynamics of the flow execution. A dynamic SWMS includes a human user in the runtime loop of a flow execution, and allows an engine to flexibly orchestrate a workflow according to the human decision and the runtime states of the environment. In this paper, we present our research in an ongoing project: Virtual Laboratory for e-Science (VL-e). An agent based solution is proposed to enhance an existing Grid enabled Problem Solving Environment framework called VLAM-G. The intelligence for problem solving strategies and for workflow orchestration is encapsulated as knowledge in two types of agents: study managers and scenario conductors.