Steel structures have been widely utilized in building structures due to their characteristic of excellent mechanical properties, lightweight, and rapid construction speed. Despite their widespread use, the design process of steel structures is intricate and time-consuming as it involves considerations such as strength, stability, and various other requirements. In recent years, with the rapid development of artificial intelligence technology, a new perspective and methodology for the design of building structures have emerged, i.e., simulating structural engineers through deep neural networks (DNN), and training the DNN to acquire domain knowledge in structural design. Therefore, based on the characteristics of steel structures, this paper models the design process of H-shaped steel columns as a deep reinforcement learning (DRL) paradigm, and proposes an innovative automated design method. This method involves the interaction between an agent and an environment based on the design codes for steel columns, and a reward function tailored to the characteristics of steel columns is designed to guide the agent in learning the design rules and techniques for H-shaped steel columns during the interaction process. Through neural network training and numerical experiments, it is demonstrated that the proposed DRL method can rapidly achieve the design of H-shaped steel columns while ensuring compliance with safety and economic criteria.