Lesson 3 / 7 · ~10 minutes
Iterate in small steps
The first version is never the last — and that's fine. The difference between frustration and smooth work isn't the AI, it's the quality of your feedback. Five rules that will save you hours:
- One change at a time. Add → test → continue. Five changes at once means that when something breaks, you don't know which one caused it.
- What I see + what I expect. Instead of "it doesn't work", write: "When I click the button, nothing happens. I expect the result to appear."
- Attach evidence. Copy the entire error message; for visual problems, attach a screenshot. Claude reads more out of them than you'd expect.
- Number things. Multiple remarks? A numbered list — one item, one problem. Nothing gets lost.
- Going in circles? Stop. A third failed attempt at the same thing = tell it to go back to the last working version and propose a different approach.
Good feedback, live
We continue with the margin calculator from lesson 2 — adding value persistence. Watch what a bug report looks like when Claude fixes the problem on the first try.
Decision game: what do you do?
Three situations you'll meet in your first week of vibecoding. Pick the best reaction.
1. After the last change the app shows a blank page. What do you write?
Option B has everything: what I see, since when, and evidence (the error message). From a report like that, Claude usually fixes the problem on the first try.
2. You have five improvements in mind. How do you brief them?
Small steps win: when something breaks, you know exactly which change caused it. (More experienced users can handle a numbered list at once — but they test after each item.)
3. Claude is "fixing" the same bug for the third time and it keeps coming back. Now what?
Repeating the same thing and expecting a different result doesn't work with AI either. Returning to a working version + changing approach is the professional reflex — and the "last working version" is what the whole next lesson is about.
Summary
- One change at a time, verify after each.
- Feedback = what I see + what I expect + evidence (message, screenshot).
- Multiple remarks → a numbered list.
- Three failed attempts → stop, roll back, different approach.