This paper proposes an evolutionary framework where a network service is created from a group of autonomous agents that interact and evolve. Agents in our framework are capable of autonomous actions such as replication, migration and death. An evolutionary mechanism is designed using genetic algorithms in order to evolve their behavior over generations. A simulation study is carried out to demonstrate the ability of the evolutionary mechanism to improve the network service performance (e.g., response time) in a decentralized and self-organized manner. This paper describes the evolutionary mechanism, its design and implementation, and evaluates it through simulations