On March 31, CITIC Securities published a research report highlighting that the current constraints of the Scaling Law have led to a gradual bottleneck in the marginal benefits of pre-trained large models. Consequently, the advancement of more sophisticated Agents emerges as a pivotal direction for large model manufacturers. The report delves into the evolution of Agents' technological framework, which has undergone three distinct stages: initially integrating workflow frameworks with pre-trained models, subsequently introducing inference models, and ultimately evolving into an end-to-end Agent framework that seamlessly integrates inference models. With the introduction of third-generation Agents, exemplified by OpenAI Deep Research, the immense potential of Agents and the accelerating impact of the Scaling Law are becoming increasingly evident. CITIC Securities anticipates that, leveraging the robust capabilities of domestic inference models, China's third-generation Agents will exhibit robust competitiveness in the market in the near future.