Today I was tinkering with a new website and learned a crucial lesson about using AI. During the process, I kept encountering some issues, and some were related to AI code, which left me puzzled as to why AI, being so precise, could still make mistakes. The reason is that AI operates in an idealized sandbox, and once it needs to interact with the outside world (via API), it has to contend with external rules, leading to problems in the original code. Additionally, multitasking and high frequency can overwhelm AI; the instructions need to be detailed and specific enough. How to use it better: 1. Learn to use specific instructions rather than vague ones. For example, instead of just saying "implement sound alerts," say: "Please implement sound alerts and include try...catch logic. If playback fails due to browser policies, print the error reason in the console and prominently display 'Click here to activate audio' in the UI." 2. Provide specific paths within vague terms. For example: "Due to certain restrictions, please do not attempt direct front-end requests, but instead write a backend route using Puppeteer or RSSHub to relay the data." 3. Break down multitasking; do not give too many tasks at once. 4. Instructions 1 and 2 can be further detailed with the help of another AI.