In this article we propose a novel approach to the evolution of agent-programs by
means of natural selection. The existing approaches (e.g. genetic programming) are
usually constrained to relatively simple program-structures since they need explicit
representation of fitness, genetic operators, and selection mechanism. We propose
a methodology that overcomes these issues by introducing a lifecycle of agents, and
their phenotype-phenotype interaction. As a consequence, an emerging evolutionary
optimization process called “natural selection” arises, which enables the evolution
of arbitrary agent-programs. Several interesting experiments are presented.