In an interview with Ryo Lu, the design lead at Cursor, Peter Yang particularly resonated with two points: 1. In AI-native companies, the boundaries of roles become very blurred. At Cursor, the division of labor among designers, product managers, and engineers is not the traditional way of each managing their own area. Ryo stated that everyone works according to their strengths; whoever is better at something takes on more of that responsibility, and then uses AI Agents to connect these tasks, creating a complete product experience. 2. The stronger the model, the more important the specifications become, rather than less important. Ryo predicts that as large language models become more capable, they will become very adept at "strictly implementing requirements according to specifications." This means that how clearly, specifically, and accurately you write your specs largely determines the quality of what the AI produces for you. 3. Release in batches, refining and adjusting based on feedback each round. Ryo discussed the internal release rhythm at Cursor: First, it is released to internal employees; Second, the latest nightly build is sent to Cursor's loyal users; Third, it is gradually pushed to regular users; Finally, it reaches enterprise users. Each batch of users is an opportunity for refinement, allowing them to discover issues and fine-tune details in a small scope before expanding.