Simple explanation

At some point AI will break something. It may change a screen you never asked it to touch, delete something, or fix one thing and break three others. This is normal. The mistake is panicking and prompting more — more prompts on top of a bad change make the mess harder to untangle.

What to do

Follow this 3-step recovery process every time: Step 1 — Stop Do not keep asking AI to fix the broken version. More prompts on top of a bad change can make the mess harder to untangle. Stop first. Step 2 — Go back to the last working save This is why GitHub matters. Your last good version should be saved before every major change. Open GitHub Desktop. Switch back to your last working branch or commit. Nothing is lost. Step 3 — Rewrite the prompt Ask yourself: what did I forget to say? What did AI misunderstand? What should it not have touched? What exact result did I want? Then write the task again with: Context. One task. Exact behaviour. Boundaries. What not to say: Do not say "that is wrong, fix it." AI does not know what right means unless you explain it. Say what changed, what should happen instead, and what must stay untouched. AI mistakes are not the end. They are feedback. They show you where your instruction was too loose.

Copy-paste prompt

I asked AI to [describe what you asked] and it built something wrong. Here is what I got: [describe what AI built]. Here is what I actually wanted: [describe what you wanted]. Help me rewrite my original prompt using the 4-part structure — context, one task, exact behaviour, and boundaries — so the result matches what I actually meant.

Course note

The course teaches prompt recovery as a core skill — not just how to write good prompts the first time, but exactly what to do when a session goes off track.

Key takeaway

Stop. Roll back to GitHub. Rewrite the prompt. In that order, every time.