Recently, I have been using Claude Code intensively and have found that the current form of AI programming tools is still too basic. Although the efficiency of AI coding has greatly improved, I still need to act as a supervisor to supervise the review of AI work, and manpower has clearly become a bottleneck in the entire process. Try to completely let go and not get involved, let the two AI review each other's work. Sometimes it is discovered that an AI claims to have completed something, but in reality, it did not or did it with numerous flaws, while another AI provides cover for it, claiming that it did a great job. Therefore, I need to constantly repeat the cycle of supervising AI to write code, identifying and correcting problems, another AI using the code, and the first AI modifying the code again, in order to continuously converge the error rate. The ideal situation would be for me to be the CEO of the company (my main job is not to participate in coding), responsible only for setting goals and frameworks, setting up security sandboxes, and letting the AI continuously iterate and improve its work until my goals are met. At most, I only need to come over every day to check the progress. However, I believe that this day will come soon, perhaps in just 1 to 2 years, and in the future, there will be many one person companies and AI employees completing all the work. Next, I will practice this strategy in my company.