A score of 42/100 in the "Understanding AI" report means you've truly entered the world of AI programming. Expanding on the experience while chatting with member @Danielshi1 Never let AI write code directly; instead, let it generate a report first, analyzing three layers of data. First layer: Reality Check – “Is it trying to fool me?” Determine whether it's a genuine database query or hard-coded fake data; whether it's a genuine API call or a simulated response. Even AIs like GPT-5 or Claude 4.5 have the flaws of "illusion" or "laziness." Sometimes, to satisfy you, it will write a piece of "pseudo-code" that looks very much like code. This is called Operational Risk control. This is the minimum requirement. If this layer fails, the system will be a disaster site upon deployment. The second layer: Code Aesthetics – "Can this be changed later?" Whether the code is clean or not. This really requires "reading more," which is what is known as "a feel for language" or "taste." AI-written code sometimes resembles a "mountain of excrement." While the functionality is achieved, variable names are often haphazard (e.g., var a, b, c), or all the logic is crammed into a single function of several hundred lines. This is called Technical Debt management. Without this level of understanding, your project is "one-off." It might work today, but next week you might want to add a new feature and find yourself completely lost, forced to start all over again. The third layer: Product Solution – "Is this really what the user needs?" Is the product design scheme reasonable? This is AI's weakest point right now, and also your most irreplaceable aspect. AI is a "Yes Man": If you ask it to create a function that "displays the password in plain text on the screen after the user enters the password", it will write it for you immediately. The code may be very elegant (second level) and the logic may be correct (first level). But you are the "gatekeeper": You need to be able to immediately recognize, "No, this is counterintuitive in terms of product logic and poses a security risk." Or, "This process is too cumbersome; users have to click three times to see the result. Can it be changed to a one-step process?" This is called Strategic Alignment. You are using business logic and user experience to constrain the technical implementation. To put it bluntly, the "contractor" doesn't need to "move bricks," but he still needs to make sure the wall is built correctly and that the cement grade is accurate. You might even have to confront the AI, challenge it, question it, and demand concrete evidence for every conclusion it draws, down to the line number of lines of code. ---Advertisement Break--- I recently started a knowledge-sharing community called "Hai Ge and His Friends" - (link in the comments). If you are - "Product manager/product owner who wants to learn AI programming and also wants to do some development work on the side." - For programmers who don't have in-depth experience with AI, this course aims to help you understand how frontline AI programmers view the functionality of each AI tool and get started quickly. - "A beginner who wants to learn Vibe coding wants to study how to create a usable project and learn the new programming paradigm faster." Then my planet might be of great help to you. In addition, the first 100 people will receive a free 1-on-1 remote AI chat with Hai Ge, and previous members even requested to extend the chat time after finishing.
