In this talk we will present a novel approach to improve the quality of selfish scheduling. We assume that not only are the tasks but the machines managed by selfish agents, and each machine agent could reject a job assigned on it in order to attract and serve a bigger job so that he/she could get more profit. We prove that for the objective of minimizing the makespan on identical machines the price of anarchy (PoA) is less than or equal to 3/2, which is strictly smaller than the PoA of the classical model. We also give a low bound 4/3 on the PoA of our model. In the end, we will discuss how to narrow the 1/6 gap.