New methods for team optimal decision, based on a stochastic agent coordination, are presented, implemented and tested. The methods are extensions of the cross-entropy algorithm (CE), initially dedicated to rare-event simulation. In particular, this paper investigates how the contributions of the agents could be involved in the simulation process. The approaches are tested on a SD-assignment problem, where each agent is considered as a 2D-assignment process. Comparative tests are made and the synergism of the system, agents and coordination, is illustrated.