Why Doing It All in One Breath Is Actually Slower
If you ask the AI in one shot to "build the whole site with pages, styles, and data," it outputs a big blob. Once something goes wrong in the middle, you can hardly point out which step is wrong—you can only scrap the whole blob and start over.
Small-step iteration = each time have it do one small piece, confirm it's fine, then do the next. The error is locked in one small step, easy to locate and fix.
Skeleton First, Details Later
The right order is coarse before fine:
1. First have it build a shell (skeleton) that opens and shows the main blocks;
2. Then fill in the core feature (the buttons actually work);
3. Only at the end deal with style, colors, animations and other details.
第1步：做出空白主页，有「输入框 + 按钮 + 列表」三个占位
第2步：让按钮能把输入加进列表
第3步：让总额会累加
第4步：再调颜色、间距、字体Verify Each Step Before Moving On
The AI saying "done" doesn't mean it's actually done. After each step, you should do a minimal verification: open it, click that button, confirm the result is right.
Verification doesn't have to be complex: "I clicked add—did it show up in the list? Is the total right?" If yes, continue; if not, have it fix this step first.
自测 · 学完检查一下
想真正动手做题、记进度、攒连胜？到互动课里练。
Compared with "having the AI do it all at once," what's the biggest benefit of small-step iteration?
答案：When an error occurs, it's easy to locate which step is wrong and fix it
Changing one small piece each time locks the error in that step, making it easier to locate and fix; one big change makes the problem hard to pinpoint.
"Skeleton before details" means: first build the ______ that opens and shows the main blocks, then fill in the core feature, and only at the end tweak the style.
答案：skeleton
Build the skeleton (shell/frame) first to get the structure working, then add features, and tweak style last—the recommended order to reduce rework.
Following "skeleton before details," which should be done last?
答案：Adjusting colors and animations
Style, colors, and animations are details and should come after both the skeleton and the core feature work.
Judge: when the AI replies "all done," you can safely move to the next step without opening it to verify.
答案：Incorrect
"Said it's done" isn't "is done." Each step should get a minimal verification (open, click, check the result), and only continue once confirmed.
After finishing "make the button add input to the list," what's the most fitting minimal verification?
答案：Actually type a value, click the button, and see whether the list gains a row
A minimal verification is to personally trigger this step's behavior and check whether the result matches—low cost and catches errors in time.
Judge: continuing to have the AI change the next step without verifying can make errors pile up and snowball into a mess.
答案：Correct
Later changes are built on unverified, possibly already-broken code, stacking errors layer by layer into an untangleable mess. So verify each step.