I often get asked by people how to use AI to create visually appealing product designs that don't suffer from AI failures. It's not an easy question to answer, as I'm not a professional myself. I can only offer my perspective as a product engineer, discussing how to create products that are both aesthetically pleasing and easy to use. First, it's crucial to understand that even a beautiful design can't save a product that doesn't solve a problem. Design here is merely an enhancement, not the deciding factor. If I had to prioritize products, it would be: problem-solving > ease of use > aesthetic appeal—like solving hunger first, then food, and ideally, good nutrition. I won't elaborate on problem-solving here, as it largely depends on the product itself. It addresses a specific need—someone needs it, there's a scenario requiring it, and your product happens to provide a solution for people or agents to use and achieve their goals. The scope of the problem can vary. So, how do you create an easy-to-use and aesthetically pleasing product? Well, you can follow this guide, or you can give the following long paragraph to your agent to review your product's codebase. 1. Put yourself in the shoes of a complete novice with no technical knowledge. If they receive our product, can they complete the main workflow and produce what they need? Identify any bottlenecks, unclear points, or inconsistencies in the product's flow, and fix them all. The goal is for users to be able to use the product without the manual. If this is impossible, provide supplementary instructions. 2. All copy must conform to the usage habits of the target language's native speakers. Carefully consider word choice; avoid dry translations. Find the best translations and explanations for similar terms from Apple, Windows systems, and well-known websites. Make the copy as simple, easy to understand, and clear as possible for local users. 3. Avoid adding additional text explanations where the product itself can clearly express the meaning. Don't be verbose in your explanations. Carefully consider whether each explanatory text is clear and concise. If the interactive visuals can clearly convey the message, don't repeat it in text. 4. When expressing numbers, ensure users understand what the data refers to and prioritize readability. For example, units like money, metrics, and distance should adhere to best practices. Data accuracy is paramount; don't skip data validation to save time. 5. Avoid repeatedly asking users for confirmation before performing actions. Only ask again if there are irreversible consequences. Prevent screens filled with confirmation prompts. Avoid large areas of red for risky actions. Consider the user experience; users aren't stupid, so don't worry excessively about them making mistakes. 6. Visually, use only one accent color per page. Avoid orange, terracotta, and the old-fashioned bright blue often seen in backend systems. Use warm gray tones, avoiding cold, harsh blue-gray. If suitable colors are unavailable, consider traditional Chinese colors. 7. Minimize the number of font sizes. Control a maximum of four different font sizes per page. Consider the overall reading rhythm, clarity, and text contrast to ensure accessibility. This includes adjusting line spacing, character spacing, and paragraph spacing as needed, while maintaining overall consistency. 8. Minimize the use of italics, monospace fonts, and all-caps. Provide detailed justification for any use of these fonts. Avoid italics in explanatory text. Use monospace fonts only for code, paths, and metrics. Avoid capitalizing text in buttons and labels. Minimize the use of periods (·) to connect content. Use only one verb for each action throughout the product. Avoid verbs that pressure users to perform actions; for example, avoid using marketing phrases like "buy now." 9. Maintain consistent design standards for buttons, spacing, rounded corners, dividers, pop-ups, animations, and shadows. Keep these elements restrained and avoid disrupting the user experience. Nested rounded corners should be concentric; the inner rounded corner should equal the outer rounded corner minus the spacing for a better aesthetic. The page defaults to a flat layout, without overlapping shadows, glows, or frosted glass. Shadows are only applied to pop-ups that truly float. Avoid nested cards. 10. In multilingual environments, prioritize the layout of the longest text, prioritizing expressiveness over aesthetics. Text can be modified, but avoid changing font size or truncation. Pay special attention to consistent display across different screen sizes, considering commonly used devices. Test all options and fix any issues. 11. For designs that repeatedly fail to improve or remain unsatisfactory after multiple rounds of revisions, have the agent provide three of their best options for you to choose from. Select the best one and guide them to optimize it until you are satisfied. I prefer to have AI generate before-and-after comparison images in a single PDF for a better reading and comparison experience. That's about it. But another very important thing is your own patience and repeated adjustments. I'm a bit of a perfectionist, and I often discuss a copy, a color, a design, or an AI more than ten times until I'm satisfied. Although AI is very smart, you often have to demand more from it and you need to invest more time and patience to achieve satisfactory results. Or, if you spend more time looking at good designs, you can actually create a product that is easy to use and not ugly.