In the air transportation system, airports are connected by companies operating flights in different ways. One is to use a shuttle system, with a high number of evenly scheduled daily flights. In such a case, a delay occurring will be smoothed out using the buffer effect of the remaining rotating flights. On the other hand, companies operating a single rotation will only be able to mitigate the delay using the turnaround duration. It is the same for transit flights. In existing simulation environments, such a fine structure is not taken into account, even if some tools were designed in the context of rail transportation. The purpose of the present work is to provide the air transportation community with an advanced delay propagation simulator using an adaptive multi-agent model. The main contribution is the introduction of a learning procedure mimicking the airlines behavior as a response to a delay. The mitigation effects will thus be made dependent on the rotation typology on the various legs 1 of the network and will adapt according to the outcomes of a decision. Furthermore, a priori knowledge about a specific airline behavior (low cost, regular) may be encoded in the learning algorithm to closely adhere to reality.