From ancient times onwards government bureaucracies have used technology. Their operations shaped the technologies but, in turn, the technologies also reshaped the operations. The massive use of AI will again reshape the organizational processes of large-scale government bureaucracies. Since these processes usually follow the law an analysis of AI driven large-scale government bureaucracies should take into account how AI affects the law. This chapter builds on a configuration approach that connects technology, the basic characteristics of the organization, and law. The configuration approach is first applied to understand how street-level bureaucracies transformed into screen-level bureaucracies and then into system-level bureaucracies. In each stage of the development of large-scale government bureaucracies the technology, the characteristics of the organization, and law constitute a stable configuration. The use of AI is still in its early stages but it is possible to explore the stable configuration that AI may trigger. This configuration is labeled as a learning-loop bureaucracy in which the individual decisions (such as welfare benefits or construction permits) result from man-machine interaction in a continuous learning-loop cycle. In order to constitutionalize these learning loops we may expect a new type of legislation that regulates the learning processes instead of the individual decisions.