Predict before you run
Artifact: one written prediction, made before the result was revealed
1. The habit
Before you press run, before you reveal the answer, before you let the AI finish its sentence — predict.
Out loud. In writing. To yourself. Just commit to an answer:
- What will this code print?
- Which test is going to fail?
- What's the bug?
- What is the AI about to suggest?
- What number will come back from this query?
Make the bet. Then look.
2. Why it works
Prediction forces retrieval. It pulls whatever model you have of the system out of your head and into the open, where reality can confirm or contradict it.
Without prediction, your model stays inside your head, in vague form, unchecked. You can hold a wrong model for years and not know it. With prediction, you find out in the next thirty seconds.
The gap between your prediction and the truth is the only thing on the page worth thinking about. Everything that matches your prediction was already known. Everything that doesn't is a free upgrade to your mental model — but only if you noticed it.
3. Concrete, small, today
Try one of these on your next session:
- Before pressing run on a script: write the expected output as a comment above the line.
- Before reading the next Stack Overflow answer: write what you think it'll say.
- Before letting Copilot complete a function: type the signature, then write the first line as a comment of what the body should do.
- Before reading a PR review: write what you think the reviewer will catch.
Tiny. Five seconds each. Compounds over a year into the difference between a junior and a senior.