This paper develops and evaluates a new decision theoretic framework in which autonomous agents can make rational choices about coordinating their actions. The framework covers the decisions that are involved in determining when and how to coordinate, when to respond to requests for coordination and when it is profitable to drop contracts in order to exploit better opportunities. Our motivating hypothesis is that enabling agents to dynamically set and re-assess both their degree of commitment to one another and the sanctions for decommitment according to their prevailing circumstances will make coordination more effective. This hypothesis is evaluated, empirically, in a grid-world scenario, taking into account three levels of commitments (total, partial and loose) and three kinds of sactions (fixed, partially sanctioned and sunk cost).