Agents that act as information brokers in large distributed systems (such as the internet) lower the cost of obtaining information. Agents have direct access to only a small part of such systems at any one time. This paper investigates the conditions in which agents successfully go through other agents in order to establish direct communication. A simulated, large information market enables agents to acquire the information they seek if and only if market shares among broker agents conform to a power law distribution. This result, if it turns out to apply more generally, implies that agency theory based on utility maximisation does not provide any guidance to system properties. Thus certain aspects of the agent decision making process are implied by the properties of the system.