Software systems become more and more complex thus the application of self-developing distributed and decentralized processing is indispensable. The complexity of such systems requires new tools for designing, programming and debugging processes which implies the fact that new approaches to decentralization should be undertaken. An idea of autonomous agents arises as an extension to the object and process concepts. The active agent is invented as a basic element of which distributed and decentralized systems can be built. The use of evolution strategies in design of multi-agent systems reveals new posibilities of developing complex software systems. The evolution plays also a key role in creation and organization of social structures. In this paper a new technology of designing and building agent systems based on genetic methods and a draft concept of a model-based approach to such systems are described. Also an application of this technology to a self-developing prediction system is presented and results of simulation experiments carried out with the use of 0-1 random time series are discussed.