The theory of multi-agent systems is considered as one of main drivers of computer networks and their ubiquity development. It is based on the implementable presumption that an agent as special software module is possible to communicate with other agents and to make reasonable decisions based on his accessible information. The implementation of multi-agent systems is realized by means of special tools which establish their architecture and standards of their interaction and communication processes. The applications of multi-agent systems are widespread both in modeling distributed systems, and as applications in computer network middleware. The preference of multi-agent systems over traditional modeling methods is that those allow taking into account the impact of individual behaviours of system components in micro level upon the behaviour and characteristics of whole system in macro level. It is attempts modeling often tending to combine with system development in computer networks. In the applications of computer network multi-agent systems often play a part of middleware which essential expand possibilities of traditional middleware, for example CORBA. This consideration is especially essential for development flexible middleware in widespread mobile devices which are getting as the essential part of computer networks. It is shortly considered such examples of multi-agent applications as architecture of perspective thread preventing system and BitTorrent data distribution system. There are considered results of modeling for the last one.