AI 做的設計，你能看出哪裡有問題嗎？

--- I gave the same e-commerce brief to Google Stitch, Figma, and Claude Design. Ten minutes later, three polished-looking design mockups appeared on my screen — yet every single one gave me a strange sense of déjà vu. The problem isn't that AI isn't good enough. The problem is that we don't have the language to say what's wrong. In this episode I'm not debating "do designers have a future." I want to ask a more pressing question: do we have the ability to spot what's wrong with the designs AI gives us? Good design has no single right answer, but bad design has standard warning signals. Seventy years of design research have turned those signals into quantifiable tools. This episode strings them all together. --- 00:00 The experiment: having 3 AIs design the same e-commerce site 00:33 The core question: can you spot what's wrong with AI-generated designs? 00:40 Google Stitch, Figma, and Claude Design — pros and cons of each 01:33 The real problem isn't the AI — it's us 01:50 Why AI always gives you the "average answer" 02:53 Good design has no right answer, but bad design has standard signals 03:14 Breaking design into 2 dimensions and 4 measurable questions 03:42 Tool 1: Where does the eye look first? DeepGaze IIE eye-movement prediction 04:37 Tool 2: Seen it — but understood it? Flesch-Kincaid + WCAG 05:29 Tool 3: Understood it — but will they act? AI Agent simulating users 06:07 Tool 4: Does the whole thing look consistent? Playwright reading CSS 07:01 Connecting all four tools into a closed loop and having Claude Code generate a fix plan 07:43 The "average face" experiment: tools give you a checklist, not an answer 08:38 The three layers of design work — execution, evaluation, definition — how far can AI go? 09:30 The more people outsource design to AI, the more valuable people with taste become --- #AIDesign #Designer #ClaudeCode #Figma #GoogleStitch #UX #WebDesign #ArtificialIntelligence #UIUX #DesignThinking #AItools #Tech #Taste #ProductDesign #AccessibilityDesign