Cost-effective is usually used as a unique standard in multi-agent task allocation process, which leads to large time consumption and low resource utilization. Therefore, the concept of task readiness is proposed, which combines cost-effective and the resource feature of the agent. According to the astringency and timeliness in the task allocation process, the learning automata algorithm is used to dynamically adjust the weights of each item of task readiness. Simulations of task allocation are done under low, medium and high task demands by using the proposed. The results show the effectiveness of the method, and the resource redundancy is reduced at least by 20%.