The algorithms of the Ant Colony Optimization family make use of the pheromon lanes used by real ants for the communication. These represent partial solutions of several individuals which compositely represent a solution with each other.
The ant becomes an agent, all agents represents the system, the colony. An agent is comparable with the ant: in relation to the overall system, she is small and simple. Since her actions have only small influence on the overall system, her loss is negligible.
The cooperation between the agents is like the ants: every agent reads locally available markings. The Ant choose a way by probability, based on the length and the pheromons on the way. Accordingly, the amount of virtual pheromones gives information about the quality of the solution.
The evaporation can be modeled, that adaptive behavior becomes achieved - the presumably biggest feature of the ant algorithms. It is possible without reinitialization to make changes to the problem during the run.