For agent algorithms in which information obtained from the environment in a large-scale distributed system is crucial, design is not limited to the theoretical or conceptual stage of development, but is a continuous process that spans into an experimental implementation, testing and comparison process. Procedures that provide information services need to be carefully matched, and perhaps customised. The performance of an algorithm must be examined in combination with the services it uses. Possible designs must be compared in the intended application setting, and improvements in effectiveness weighed against increased cost. The AgentScope project explores how to enable the experimental stages of multi-agents systems development.