First, we present a new computational market theory, implemented in the HOMEBOTS system. It shows how large numbers of electrical loads can be automatically managed by autonomous agents, that communicate and negotiate in an electronic multi-commodity market leading to optimal use of electrical power. The advantages of this agent-based approach compared to traditional methods for power load management are described. Second, we demonstrate through simulated business scenarios that significant energy cost savings can thus be achieved. Third, our approach has been tested in a field experiment in an energy distribution area in the South-East of Sweden. The performed field tests show that the real-time requirements for agent communication over the power line in energy services are well met in realistic application settings.