Objective
Your mission
Practice the predict-before-reveal habit 5 times and track whether your predictions get more accurate. Output: prediction_log.md.
You need: Any AI coding tool with a project you can build and break.
Time: 30 minutes.
Step 1: learn the habit (3 min)
Before pasting any error to the AI, write down your prediction:
"I think the error is because [your guess]."
One sentence. You don't need to be right. You just need to think before you reveal.
Example:
- Error:
TypeError: Cannot read properties of undefined (reading 'map') - Your prediction: "I think the data I'm trying to .map() over is undefined — maybe the API hasn't returned yet."
- AI says: "The
items variable is undefined because the API call is async and the component renders before data arrives." - Compare: Your prediction was close! You guessed the data was undefined and timing was involved.
Wrong predictions are valuable too:
- Error:
SyntaxError: Unexpected token '<' - Your prediction: "Missing closing tag in HTML template."
- AI says: "The API is returning HTML instead of JSON — the
< is from <!DOCTYPE>." - Compare: Wrong, but now you've learned:
Unexpected token '<' often means "receiving HTML when expecting JSON."
Why predictions build skill
Predicting before revealing forces your brain to engage with the problem. Even when your prediction is wrong, the comparison between your guess and the actual cause creates a learning moment. Over time, your predictions get more accurate — and that's the skill that lets you catch the AI when its diagnosis is wrong.
Step 2: build something (5 min)
Build a small app with AI, or use an existing project. Something with a form, a list, and some data — enough to generate different types of errors.
Step 3: run 5 error cycles (20 min)
For each cycle:
- Either encounter a natural error or create one intentionally (break a variable name, remove an import, change a calculation)
- Write your prediction FIRST — before pasting to AI
- Paste the error to the AI
- Compare your prediction to the AI's diagnosis
Write in your file:
## Cycle 1
Error: [error message]
My prediction: "I think the error is because..."
AI's diagnosis: [summary]
Result: [right / wrong / close]
What I learned: [specific insight]
## Cycle 2
[same structure]
## Cycle 3
[same structure]
## Cycle 4
[same structure]
## Cycle 5
[same structure]
Step 4: reflect (2 min)
Write in your file:
## Reflection
Are predictions getting more accurate across cycles? [yes/no]
Are similar errors getting easier to predict? [yes/no — which patterns are you recognizing?]
Which wrong prediction taught you the most? [specific cycle + what you learned]
You predicted a missing HTML tag, but the AI says Unexpected token '<' came from an API returning HTML instead of JSON. Was the prediction wasted?
Comparing a wrong guess with the real cause is the learning moment. Next time, that error suggests "HTML where JSON was expected."