Since joining OpenAI, Peter Steinberg has been committed to advancing the development of next-generation artificial intelligence agent technology. He points out that the direction of AI development lies not only in answering questions but also in invoking tools, collaborating across systems, and continuously acting within their environments, which has become the new core of industry competition. In the interview, he discussed the different application paths of OpenClaw in China and the United States: The U.S. relies on individual innovation, where technology spontaneously diffuses from grassroots programmers and hacker communities, depending on individualistic culture and decentralized political structures. Although diffusion is uneven, it leads in raw innovation. China, on the other hand, relies on platform integration, where large firms first optimize technologies and package them into tools, penetrating through infrastructure like cloud services. This depends on collectivist culture and a centrally commanded system, prioritizing scale effects. While there is less room for originality, implementation is highly efficient. To improve AI agents, he proposed measures such as improving tool documentation, clarifying tool logic, setting iteration limits, adopting a plan-execute-improve model, defining strict roles, and inserting checkpoints. Regarding the safe collaboration between personal and professional agents, he emphasized the need to implement safety rules, use allow/deny lists, establish sandbox environments, and strengthen cognitive humility through mechanisms like confidence estimation prompts and critic-validator loops. OpenAI is accelerating progress in this technical direction by investing in startups like Isara, building multi-agent collaboration platforms, optimizing model costs, and promoting technologies such as 'model distillation' and 'prompt optimization.'