Crowdsourcing system is a complex adaptive system. The method of crowdsourcing research proposed in this paper is mainly based on this property. Crowdsourcing is an emerging research field of social computing, basing on network is its main feature. Unfamiliar people could make collaborative innovation and profit sharing through the Internet by crowdsourcing. The mechanism design of the crowdsourcing platform always varies from one to another, and a good mechanism design is the basis for stable development of crowdsourcing platform. Based on the theory of complex adaptive system, this paper proposes a new way to research crowdsourcing platform. This paper takes crowdsourcing translation as the research platform, using the concept of Agent in complex adaptive system to model and simulate the roles among crowdsourcing platform. After research, the Agent models are constructed primitively. Besides, the rules of Agent state transition and interaction between the Agents are defined. At last, we establish a simple simulation framework, by dynamic simulating of multi-agent collaboration, to simulate the development of a crowdsourcing platform and verify the reasonableness of the designed mechanisms, which could provide a reliable, credible basis for the design of the crowdsourcing platform and a convenient simulation environment for the study of crowdsourcing platform.