Twelve practical lessons and four applied projects. Work through the Prepare, Practise, Check and Apply stages, keeping your own attempts visible. All supplied situations are fictional. Written work is self-assessed, not automatically graded.
Use your browser’s Print command to print or save a PDF. Make your own attempt before reading the worked reasoning.
Prepare / LESSON 1
A learning contract is a small agreement with yourself: what you will practise, what you will produce and how you will check it.
Choose a task with a useful outcome. Limit it to one session and name what you will do before consulting AI. Your contract should allow revision when you discover a missing prerequisite. It is a guide to action, not a promise to master everything today. Include a stop condition: an artifact you can inspect or a question you can now answer independently.
Instead of “learn planning,” try “make a 30-minute study plan with an unaided attempt and a final check.” The result can be inspected before you expand the scope.
Write your task, expected artifact, time limit and a visible success criterion.
Which contract makes learning visible?
Create a short plan, explain two choices and test one step. The artifact, explanation and test provide useful evidence.
Reduce your available time by half. Keep a useful outcome by narrowing the task.
Prepare / LESSON 2
Before revealing an explanation, predict a result and state why. A specific prediction gives you something to compare with evidence.
Use a number, an expected error or a proposed next step. Record the reason even if you feel uncertain. After checking, separate a correct result from correct reasoning: a lucky guess can hide a faulty rule. Avoid making predictions so vague that any outcome would count as agreement.
A learner predicts that doubling the number of identical boxes doubles the total items. They state that the items per box must remain constant, making the hidden condition visible.
Predict the result of one small example related to your goal. Name the assumption it depends on.
Your prediction was right but your reason was wrong. What next?
Repair the reasoning and test a fresh example. A new example can reveal whether the corrected rule is usable.
Change the assumption in your prediction. Explain whether the expected result still follows.
Prepare / LESSON 3
Sort your knowledge into three groups: I can do it, I can explain it with help, and I need to investigate it.
Use a short task to place each skill, rather than guessing from familiarity. Keep the map narrow: three or four parts of your current outcome are enough. The next practice should target the earliest gap that blocks useful work. Update the map after an attempt, and record the support you used so assisted and independent performance are not confused.
For a small inventory summary, a learner can add counts, can calculate an average with a reference and cannot explain a missing value. Practice starts with what a missing value means.
List three parts of your task and the evidence for your current level on each.
Which evidence best supports “I can do it”?
A fresh task completed without the model answer, with a check. This shows performance under stated conditions.
Choose one item marked “with help” and design a check with less support.
Practise / LESSON 4
A focused request lets AI respond to your actual difficulty while leaving you a meaningful next action.
Show your goal and attempt. Name the point where your reasoning stops. Ask for one question, then wait before requesting a hint. If the assistant writes a complete answer anyway, use a different example for your next unaided attempt. The learning comes from this exchange of attempts and feedback, not from finding one perfect prompt.
“Here is my checklist. I am unsure whether the order handles missing information. Ask me one question about that step before suggesting a rewrite.”
Write a coaching prompt around a real gap in your attempt.
Which request preserves a useful next thinking step?
Ask me to identify the missing condition before explaining it. A targeted question can expose the gap before the solution arrives.
Rewrite the request for a learner who knows the basics but repeatedly makes one error.
Practise / LESSON 5
Examples teach more when you explain what stayed the same and what changed between them.
Choose two small cases that differ in one important way. Predict whether the same method applies and why. Ask AI for a contrasting example if helpful, then verify it before relying on it. A pair of examples is not a universal proof, but it can reveal the boundary you should investigate next.
A list contains scores 0, 4 and 8. Another contains a blank, 4 and 8. Zero is a measured score; a blank may be unassessed. Treating both as zero changes what the average claims to describe.
Create two examples that differ in one meaningful condition. Explain what the change affects.
Why is a contrast useful?
It can reveal a condition hidden by one example. Changing one condition makes its role easier to inspect.
Add a third example that tests the boundary of your explanation.
Practise / LESSON 6
Turn feedback into a named change and another attempt. A comment is useful only if you know what to do with it.
Link the feedback to a criterion you chose in advance. Preserve the parts that are already accurate. If several issues appear, address the one that blocks the task first. Ask the reviewer to explain a vague comment with an example, then make the revision yourself. Keep a short note of why the change improves the result.
“Your summary is unclear” becomes “the reader cannot see which action you recommend; put that action in the opening sentence.” The learner revises and tests the message with a different reader.
Record one feedback comment, the criterion it concerns and your next revision.
Which response makes vague feedback useful?
Ask what part fails which criterion and request a small example. This connects feedback to an inspectable action.
Write a follow-up task that checks whether the improvement carries into a new example.
Check / LESSON 7
Match your check to the kind of claim. Calculations, factual claims and procedures need different forms of evidence.
Reproduce arithmetic from the inputs. Inspect an original source for an external factual claim. Test a procedure with a case whose expected result you know. Ask what the check can establish and what remains outside it. Repeating the assistant’s answer is not a separate source of evidence.
A stock checklist can be tested with a complete entry, a duplicate and an entry missing a quantity. A confident description of the checklist is not a test of its handling of those cases.
Select one important claim or step and describe a check you can perform.
How should you check a total calculated by AI?
Recalculate from the original values and inspect which values were included. A reproducible calculation checks the method and input coverage.
Name a limitation of your chosen check and an additional check if the consequence of error matters.
Check / LESSON 8
What happened and why it happened are different claims. Report each at the strength the evidence supports.
A comparison can show a difference while leaving its cause uncertain. Consider prior experience, time and help received before attributing the result to a single learning method. You can use an observation to plan another test without claiming a general causal rule. A small personal comparison is a clue about your practice, not a verdict for everyone.
You finish a task faster in a second session, but it is also more familiar. The observation is faster completion; familiarity and the new strategy are both possible explanations.
Write one observation from your learning and two possible explanations.
What can a faster second attempt establish by itself?
This attempt was faster; the cause needs more investigation. This separates a measured result from its possible causes.
Design a small follow-up that would help distinguish your two explanations.
Check / LESSON 9
Make confidence about a specific task and attach it to an attempt, not a feeling of familiarity.
Say what you can do, under what conditions and with what support. If evidence is thin, name the next check rather than choosing an impressive confidence label. This helps you decide where independent practice is worthwhile and where a reference is still useful. Confidence can change when the task or conditions change.
“I can identify the whole in these three percentage examples without notes” is more informative than “I am great at maths.” It makes a boundary visible.
Write a bounded statement of what you can currently do and the evidence behind it.
Which statement is best calibrated?
I solved two new examples without hints, but have not tested a different context. It reports performance and its limits.
Change the context or time limit. What new evidence would you need?
Apply / LESSON 10
Use support when it helps you progress, then remove one piece of it for the next attempt.
Decide what support is essential and what you can try without. You might hide a worked example while retaining a brief checklist, then later remove the checklist. If the task becomes unproductive, restore a small hint and identify the gap. Support fading is a practical experiment, not a rule that help is bad.
A learner first uses a model email, then a structure checklist, then writes a different update from a blank page. Each attempt records the help available.
Plan three attempts with gradually reduced support. Keep their difficulty reasonably comparable.
What should you do after needing a worked solution?
Try a fresh example with less help and record what happens. A new attempt checks what the support enabled you to learn.
Write the smallest hint you would restore if the next attempt stalls.
Apply / LESSON 11
Explain one principle with a concrete example, then ask someone to apply it in a new case.
Keep the explanation short enough that the learner must still do something. Ask for reasoning, not only agreement. An accurate answer with an incorrect explanation can reveal guessing. If the learner struggles, identify the specific misconception and give a focused prompt rather than repeating the whole explanation louder.
To teach attendance rate, explain attendees divided by registrations. Then ask for the rate when 12 of 16 attend and why 16 is the denominator. The answer is 75%, but the explanation matters too.
Write a short explanation, a new application question and the reasoning you would look for.
What is a useful teaching check?
Ask for a fresh application and an explanation of the method. This makes both the result and reasoning inspectable.
Write feedback for a plausible wrong answer to your teaching question.
Apply / LESSON 12
End with a new independent task and a realistic review plan. A completed session should make the next action clearer.
Compare your first and final attempts, noting any differences in task difficulty and support. Keep one useful correction and one unanswered question. Choose a future task that uses the skill with changed details. Your plan can include when to ask for help again. Completion is participation; capability needs repeated evidence in relevant situations.
After learning to compare two plans, a learner schedules a later comparison with a new constraint and explains which recommendation changes. This checks the reasoning instead of memorising the original preference.
Write your next task, review date, success criteria and fallback if you get stuck.
What is the best next step after completing this guide?
Try a fresh real task and record where you still need support. A fresh task connects the guide to your own capability.
Choose a different setting where the same learning cycle could be useful.
A fictional cupboard starts with 12 notebooks, receives 8 and gives out 5. A draft summary reports 25 remaining. Produce a checked summary and a simple checking routine.
Starting stock: 12. Received: 8. Distributed: 5. No other changes are supplied. All data is fictional.
A short stock summary, a reproducible calculation and a test with different numbers.
Distinguish incoming from outgoing items. Use only supplied changes. Explain the calculation.
Remaining stock is 12 + 8 − 5 = 15. The draft added distributed items instead of subtracting them. A useful check labels each change as incoming or outgoing before calculating.
A new day starts with 15, receives 3 and distributes 7. Calculate the new total and explain which sign belongs to each change.
Record your first attempt, outcome, artifact, verification, revision, changed-situation response, explanation without notes and next independent task.
You have 30 minutes to learn how to write a clear meeting action item. You need to produce an example that names the owner, action and deadline.
30 minutes available. You have one checked sample and a blank page. AI support is optional. No real names or private meeting notes are needed.
A 30-minute plan, one action item and an independent example with changed details.
Time totals 30 minutes. Include a first attempt, feedback and a new unaided example. Every action item identifies who does what by when.
Use 4 minutes for a first attempt, 6 to compare with the sample, 8 for focused practice, 7 for a fresh independent example and 5 for reflection. Example: “Jordan will send the revised agenda to the group by Thursday at noon.”
You now have only 18 minutes. Preserve an attempt and independent check while reducing scope.
Record your first attempt, outcome, artifact, verification, revision, changed-situation response, explanation without notes and next independent task.
A fictional club asks new members to label shared folders. The draft says “use a sensible name.” Rewrite it into a small procedure with an example and a check.
For this exercise, the agreed pattern is YYYY-MM-DD-topic. Topics use lowercase words separated by hyphens. No actual private files need to be renamed.
A short instruction, a valid example, an invalid example and a checking question.
State the agreed pattern explicitly. Explain enough to distinguish a valid example from an invalid one. Do not change real files.
“Start with the date in YYYY-MM-DD form, add a hyphen, then a short lowercase topic using hyphens between words.” Example: 2026-10-15-team-notes. “notes final” is missing the date and does not follow the agreed pattern.
A learner asks what to do when the topic contains several words. Revise the instruction so it answers this without a live explanation.
Record your first attempt, outcome, artifact, verification, revision, changed-situation response, explanation without notes and next independent task.
A fictional group must practise a presentation on Tuesday with 10 participants. Room A is available Tuesday and seats 12 but has no screen. Room B has a screen and seats 20 but is only available Wednesday. Paper practice can meet the goal.
Tuesday is fixed. Ten participants. Room A: Tuesday, 12 seats, no screen. Room B: Wednesday, 20 seats, screen. Paper is available.
A short recommendation with a constraint, a tradeoff and a condition that would change your choice.
Check feasibility before preferences. Do not invent availability. Connect the choice to the learning goal.
Room A meets the fixed date and capacity needs. Use paper for practice; the tradeoff is losing the screen demonstration. Room B becomes an option only if the date constraint changes.
The group grows to 15 while Tuesday stays fixed. Explain why your original recommendation must be revisited instead of forcing the same answer.
Record your first attempt, outcome, artifact, verification, revision, changed-situation response, explanation without notes and next independent task.