Break It, Then Fix It
Module 3 · Under the Hood

Break It, Then Fix It

Learning by doing: cause a real error on purpose, hand it back to the AI, and watch it get resolved — twice.

🛠️ Learning by Doing

You’re going to break your app on purpose, twice, and fix it both times. This removes the fear factor before a real error ever surprises you.

💻 Code Lab: Break #1 — Ask for Something Contradictory

Give the AI a request that doesn’t quite make sense:

“Make the fun-fact button show a new fact only when I have not clicked it before, but also make it work every single time I click it.”

This kind of contradiction often produces a bug, a broken feature, or a confused result. Run it. Click the button. Notice specifically what happens that shouldn’t.

Report it back, describing exactly what you see:

“I clicked the button and nothing happened / it showed an error / it only worked once. Here’s what I expected instead: [describe it].”

💻 Code Lab: Break #2 — Trigger a Real Error Message

Ask for something slightly beyond what’s reasonable, on purpose:

“Add a feature that fetches live weather data from an external source and displays it on the page.”

Depending on your tool, this may produce an actual error message, a broken preview, or a partial result. Whatever appears, don’t try to interpret it yourself.

Copy the exact text you see (or a screenshot, if that’s easier in your tool) and hand it back:

“I got this error: [paste it exactly]. What happened, and can you fix it or suggest something simpler that will actually work?”

Read what the AI says happened. You don’t need to fully understand every word — just confirm the fix, when applied, actually resolves what you saw.


🔍 What You Just Practiced

In both cases, you didn’t diagnose the problem yourself. You reported symptoms accurately — what you did, what you saw, what you expected instead — and let the AI do the diagnosis. That’s the entire skill this lesson is building: precise reporting, not technical interpretation.

⚠️ Common Pitfall: Trust the Fix, But Verify It

When the AI explains what went wrong, it will sound confident — AI models generally do, even when they’re not fully correct. A confident-sounding explanation is not the same as a correct one.

The fix for this isn’t reading the code yourself. It’s the same Test step from Module 2: run it again and check the specific thing that was broken. If it now works the way you expected, the explanation doesn’t matter much. If it still doesn’t, report that too — “still broken, here’s what I see now” is a completely valid next message.


⏭️ Up Next

Time for the lecture: how to actually read an error message like a detective, and why “confidently wrong” is a real pattern worth watching for.