← Open the interactive adventure
A complete study guide with 18 chapters, six realistic missions and practical exercises for learning with AI while keeping your own thinking visible.
Use your browser’s Print command to print this guide or save a PDF. All mission situations and data are fictional. Work through decisions before reading their consequences. Written work is self-assessed; completion is not an externally assessed qualification.
Launch / CHAPTER 1
Choose something you want to be able to do in a real situation. A topic tells you where to look; an outcome tells you what to practise.
Name an action, a context and a quality standard. “Understand data” is hard to assess. “Explain a weekly sales table and flag one limitation” gives you a small artifact to inspect. Keep the first version achievable in one or two sessions. If the outcome needs several unfamiliar abilities, reduce the scope before increasing the time. Your goal can evolve after you have evidence from an attempt.
A volunteer wants to learn presentations. Her first outcome is a three-slide proposal that states the event purpose, compares two venues and requests a decision. She does not need to master animation before presenting a useful proposal.
Write: In [situation], I will [observable action], meeting [two criteria]. Then name the smallest version worth making.
Counting videos watched or prompts sent as proof that you can perform the task.
Launch / CHAPTER 2
Try a short version of the task before looking at a model answer. A baseline tells you where to spend your effort.
Keep the attempt and record any help you used. Separate what you know, what you can do and what you cannot yet explain. A blank start is still useful information: write your interpretation of the task and the first uncertainty. Later, use a comparable task with changed details. A polished final answer does not show improvement if it was copied or if the final task was much easier.
Before learning to write an update, a learner drafts four sentences. The draft contains background but no requested action. The next practice focuses on making the decision visible rather than improving every aspect of writing.
Make a three-minute attempt. Highlight one strength, one uncertainty and the first step where you needed help.
Treating an early mistake as a permanent trait rather than evidence about the next practice step.
Launch / CHAPTER 3
Choose two or three criteria before you ask AI to evaluate your work. Otherwise feedback may optimise for polish instead of purpose.
A useful criterion describes something visible: correct denominators, a clear requested action, a schedule that fits, or an explanation of a limitation. Avoid vague criteria such as “professional” unless you define what they mean for this audience. Weight non-negotiable requirements first. You can revise the criteria when new constraints appear, but record why they changed.
For a fictional sign-up process, quality means no duplicate confirmations, a clear response to missing details and a manual fallback. A more complicated system is not automatically better.
Write three checks someone could perform on your result. Include one accuracy check and one usability check.
Asking “Is this good?” without specifying what the work is supposed to accomplish.
Build / CHAPTER 4
A learning route should help you attempt the task, not become a substitute for attempting it.
List the decisions or actions a successful attempt requires. Identify the earliest missing prerequisite. Practise that piece with a small example, then return to the whole task. AI can propose a route, but you should remove material that does not serve your current outcome. Revisit the route after each attempt: the bottleneck may move from calculation to interpretation, or from planning to execution.
To summarise a survey, a learner needs to identify the population, calculate a percentage and explain a limitation. Styling a dashboard can wait until those three steps work.
List three parts of your task. Circle the first bottleneck and design a five-minute exercise for it.
Following a huge AI-generated syllabus before making a single useful artifact.
Build / CHAPTER 5
Tell your AI assistant how to support your thinking: ask, wait, hint and check. Give it your attempt before requesting a complete explanation.
Include your outcome, relevant context, current attempt and exact difficulty. Ask for one question at a time and a small hint before a worked solution. A prompt does not guarantee the assistant will follow this pattern. If it gives away the answer, switch to a fresh example and attempt it unaided. Use fictional or anonymised details when the skill can be learned without sharing private information.
“I am comparing two room layouts for 12 beginners. Here is my plan. Ask one question about participation before suggesting changes, and wait for my answer.” This asks for coaching around a specific criterion.
Draft a coaching request with four parts: outcome, attempt, difficulty and turn-taking rule.
Believing the perfect wording of a prompt matters more than the attempts and checks that follow.
Build / CHAPTER 6
Use enough support to keep moving, then reduce it when you try again.
A practical ladder starts with a clarifying question, then a partial step, then an analogous example, and finally a full worked solution. Record the highest level you needed. A solution can be useful instruction, but it is not evidence of independent performance. After using one, explain its steps and attempt a new example with fewer hints. Do not spend an entire session stuck simply to avoid receiving help.
If a learner cannot calculate attendance rate, first ask “What group was eligible to attend?” Then identify the denominator. An analogous example can use different numbers so the learner still completes the original calculation.
Write a question hint and a partial-step hint for your own task. Decide when you would move to a worked example.
Either demanding no help at all or immediately asking AI to finish everything.
Investigate / CHAPTER 7
Make a prediction before reading an answer. It creates a specific comparison between your model of the situation and the explanation.
A prediction can be a number, an outline, a likely failure or a next step. Add a brief reason. After revealing an explanation, identify what changed and why. If you predicted correctly, test whether your reasoning was also correct; a lucky answer can conceal a mistaken rule. Keep predictions low stakes so you can revise them freely.
A learner predicts that raising a value by 20% and lowering it by 20% returns to the start. Testing 100 → 120 → 96 reveals that the percentage base changed. The useful lesson is about the base, not memorising 96.
Predict what will happen in one small example. Reveal or calculate the result, then explain the mismatch or agreement.
Reading the answer first and then mistaking its familiarity for something you could have produced.
Investigate / CHAPTER 8
Select the claims that matter and verify them using a source or test outside the assistant’s confident wording.
Open an original source and check whether it supports the exact claim, population and timeframe. For arithmetic, reproduce the calculation. For a process, use a small test case with a known expected result. Asking the same assistant again can reveal inconsistencies but is not independent verification. Record what the check establishes and what remains uncertain.
An AI summary says 75% of customers prefer a new opening time. The source is six positive responses out of eight returned forms, while forty customers were invited. “75% of respondents” fits the evidence; a claim about all customers does not follow.
Choose one consequential claim. Write its evidence, a check you performed and a limitation that remains.
Treating a real citation as relevant without checking the passage, or treating a plausible citation as real without opening it.
Investigate / CHAPTER 9
Make a mistake useful by locating where the reasoning first went off track.
Distinguish missing knowledge from an execution slip, a hidden assumption or misunderstood instructions. Each suggests a different response. A calculation slip might need a checking routine; a wrong denominator needs conceptual practice. Ask AI for possible explanations, then compare them with your actual steps. Make a new example that would expose the same error if it returned.
A learner calculates growth from 20 to 30 as 33%. They divided the increase by the final value. Label the starting value and calculate 10/20 = 50%; then test growth from 40 to 50 without assistance.
Record: what I did, why it failed, the corrected rule and a fresh test of that rule.
Replacing your whole answer with a model answer without finding the cause of the difference.
Decide / CHAPTER 10
Facts describe what is known. Recommendations propose what to do under particular priorities. Keep the two visible.
A sound recommendation connects evidence to explicit criteria and acknowledges a tradeoff. If an input is missing, state an assumption rather than inventing it. Ask what change would reverse your preference. There may be several reasonable options, especially when values or constraints differ. The goal is defensible reasoning, not a universal winner.
A Monday launch is possible only if the optional FAQ is removed. Recommending the smaller release is a judgment about scope and timing. Saying the team has approved Monday is a different factual claim and cannot be invented.
Underline the facts in a draft recommendation. Circle assumptions and write one condition that would change your choice.
Letting persuasive AI wording silently turn a proposal into an approved decision.
Decide / CHAPTER 11
Use criteria to make tradeoffs visible. Numbers can organise a judgment, but they do not make subjective choices objective.
Start with non-negotiable constraints. Eliminate an option that cannot meet them before adding up preference scores. If you use weights, explain who chose them and why. Test whether a small change in a weight reverses the result. If it does, the decision may be sensitive enough to warrant more evidence or a reversible trial.
One workshop plan fits 45 minutes but has no learner practice. Another covers fewer ideas and includes an independent exit task. If demonstrating a skill is the goal, the second fits better even if the first contains more content.
Compare two options using three criteria. Name a tradeoff, a constraint and a reversible way to test the choice.
Adding many invented scores until an uncertain recommendation looks mathematically inevitable.
Decide / CHAPTER 12
A change in time, audience or resources reveals which parts of your plan serve the goal and which are optional.
Protect the core learning action when reducing scope. Decide what to remove, what to simplify and what must remain. Explain the tradeoff rather than silently shrinking every activity. A good adaptation may look different from the original artifact while preserving its purpose. Test that the revised version still meets the non-negotiable criteria.
A workshop starts eight minutes late. Shorten the demonstration and group reporting, but keep each learner’s attempt and the independent exit check. The plan sacrifices coverage to preserve evidence of learning.
Cut your available time by one third. Rewrite the plan and explain what you protected.
Removing practice and verification first because presentation content feels easier to keep.
Transfer / CHAPTER 13
Try the same principle with changed details before jumping to a completely different problem.
A near-transfer task preserves the core structure but changes numbers, wording or examples. Explain why the method still applies. If you fail, identify the step that depended on surface details. Use a comparable difficulty level so success or failure is informative. Then increase the distance to a new context or constraint.
After calculating a survey response rate, calculate attendance rate from registrations. Both use a part divided by a relevant whole, but the denominator answers a different question. Correct arithmetic alone is not enough.
Write one fresh example that changes the details but keeps the principle. Solve it without the original answer open.
Repeating the same example and calling the repetition proof that you can use the idea elsewhere.
Transfer / CHAPTER 14
A useful advanced task requires a choice about how to apply the idea, not merely a larger number or longer answer.
Change the context, introduce an irrelevant detail or remove an obvious cue. Ask the learner to select a method and justify it. Keep the task aligned with the skill rather than adding unrelated complexity. Compare the attempt against explicit criteria and note any hints used. Failure identifies the next practice target; it does not invalidate all previous learning.
A learner who can calculate percentages receives two survey summaries with different response rates. The harder task is to explain why the percentages alone may not establish a trend in the whole population.
Create a task where you must decide whether your usual method applies. Include an explanation of the boundary.
Equating difficulty with obscure wording, excessive workload or unrelated background knowledge.
Transfer / CHAPTER 15
Defend one important decision in your artifact using its purpose, evidence and limits.
Close the example and explain the reasoning in your own words. If you can repeat a phrase but cannot reconstruct a calculation or justify a choice, return to a smaller example. Invite one fair challenge. Answer it before requesting feedback from AI. This checks your understanding of the work, not your ability to perform a polished speech.
Asked why a report says “among respondents,” a learner explains that people who did not respond may have different views. They can state what the data supports without dismissing the data entirely.
Give a 90-second explanation containing a choice, a check and a limitation. Then answer one skeptical question.
Memorising an AI explanation that you cannot connect to the inputs or constraints of your own work.
Sustain / CHAPTER 16
A useful portfolio includes attempts and changes, not just a polished final artifact.
Save a baseline, one consequential mistake, a check, a revision and an independent application. Note the help received. These records let you distinguish improvement in the product from improvement in your own capability. Keep the evidence brief enough that you will actually maintain it. Remove private details when a fictional description shows the same skill.
An email portfolio includes the first vague draft, a note that the requested decision was missing, the revised message and a new update written unaided. The sequence makes the learning visible.
Select four pieces of evidence that show how your reasoning changed. Add one sentence explaining each.
Collecting only completion badges, prompt histories or finished outputs with no explanation of your contribution.
Sustain / CHAPTER 17
Plan a later unaided attempt that checks the skill you care about. A date is a reminder to act, not evidence that learning is permanent.
Choose a manageable task, a review date and success criteria. Use a new example and record whether you needed support. The right interval depends on the task and your goals; this tool does not claim an optimal schedule. If recall is weak, use feedback and try again. If recall is strong but application fails, change the practice instead of merely rereading.
A learner revisits a week later by writing a different decision email in ten minutes. They check the requested action, deadline and separation of fact from recommendation. Any gap becomes the next focused exercise.
Choose a review task and date. State what you will do if it is too easy and what support you will use if it is too hard.
Scheduling rereading only, without checking what you can produce when the original answer is hidden.
Sustain / CHAPTER 18
Bring the pieces together in a small project: frame, attempt, get support, verify, transfer and reflect.
Choose a useful artifact and keep the cycle short enough to finish. Use AI where it improves feedback or supplies a helpful example, then remove it for an independent check. Revisit your initial goal and criteria before judging the result. A completed cycle does not make you an expert in every topic, but it gives you a repeatable way to investigate your next learning challenge.
A club organiser redesigns sign-ups using fictional cases. They map the existing process, identify duplicates, test a proposed workflow, handle a session change and explain the remaining manual step. Their final artifact is useful because its assumptions and failure cases are visible.
Complete the capstone brief in this tool. Attach your reasoning as text, record a test and define your next independent project.
Treating the tool’s completion state as an externally assessed credential or stopping before trying the skill in a new situation.
MISSION 1 / Evidence investigator
A fictional community library is deciding whether to repeat a beginner session. You must turn a confident AI summary into a useful, limited recommendation.
A short evidence brief with labelled calculations, a limitation and a next-step recommendation.
Use the correct denominator. Keep claims within the data. Separate your recommendation from the facts.
Forty people were invited. Twenty attended. Eight returned forms. Six of those eight found the session useful. AI writes: “75% of the community found it useful.”
What do you do first?
Write your choice and reason before reading the consequences below.
Practice: Calculate attendance rate, response rate among attendees and usefulness among respondents. Label the denominator each time.
A second session has 30 attendees and 12 responses. Nine respondents rate it useful. An AI assistant says satisfaction increased because nine is more than six.
Which comparison is defensible?
Write your choice and reason before reading the consequences below.
Practice: Write two sentences comparing the sessions without claiming a population-wide trend.
The organiser can run one small follow-up or commit to a year of sessions. Six respondents from the second session requested a beginner follow-up.
What is your recommendation?
Write your choice and reason before reading the consequences below.
Practice: Write your recommendation, one limitation and one question the pilot should answer.
A third session receives 18 positive responses out of 30 responses, with 40 attendees. Explain how usefulness among respondents and response coverage changed.
Check 60% usefulness and 75% response coverage. Explain that a change among respondents is not automatically the same as a change in everyone’s view.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 2 / Communication lead
Your fictional team has a draft ready, a delayed review and a deadline. Write an update that helps someone make a decision without inventing agreement.
An email under 150 words with two options, a recommendation and a requested decision.
Preserve the actual status. State tradeoffs. Make the action and deadline visible.
The draft is ready. Review needs two working days. A Monday launch requires dropping an optional FAQ. Full scope moves launch to Wednesday. AI writes “Everyone approved Monday.”
How do you handle the draft?
Write your choice and reason before reading the consequences below.
Practice: Write a factual opening sentence and a one-sentence decision request.
A stakeholder says the FAQ is essential for first-time users. Your earlier preference was the smaller Monday launch.
What changes?
Write your choice and reason before reading the consequences below.
Practice: Recommend a next step while acknowledging the new requirement and any assumption that needs confirmation.
Your update is 400 words long. The request appears in the final sentence. Your reader has one minute.
Which revision best serves the reader?
Write your choice and reason before reading the consequences below.
Practice: Write the final email in fewer than 150 words. Include a subject, options, recommendation and deadline.
The decision-maker rejects Wednesday but confirms the FAQ is essential. Write a clarifying question instead of inventing a feasible plan.
Identify the conflicting scope and timing constraints. Ask what resource, deadline or requirement can change; do not promise work capacity that was never supplied.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 3 / Learning designer
Design a 45-minute session for twelve beginners to practise writing an AI coaching prompt. One shared screen and paper are available; individual AI accounts are not guaranteed.
A timed agenda, paper-based fallback and an independent exit task.
Every learner attempts the skill. Time fits. Feedback uses criteria. The plan survives an internet failure.
AI proposes 35 minutes of slides followed by questions. The outcome is for every learner to write and revise a coaching prompt.
Which plan better matches the outcome?
Write your choice and reason before reading the consequences below.
Practice: Allocate 45 minutes across demonstration, first attempt, feedback, revision and an exit task.
The internet connection fails. Several learners do not have accounts.
What is the strongest fallback?
Write your choice and reason before reading the consequences below.
Practice: Write a paper activity with instructions and three feedback criteria.
The room becomes available eight minutes late, leaving 37 minutes.
What do you shorten first?
Write your choice and reason before reading the consequences below.
Practice: Produce a 37-minute agenda and a new exit task that does not copy the demonstration.
The next group has 24 learners and the same 37 minutes. Adapt the participation structure while keeping an individual artifact from everyone.
Use pairs or small groups for feedback, but retain each learner’s written attempt and exit response. Do not assume twice as many people can each report to the whole room.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 4 / Workflow explorer
A fictional club tracks event sign-ups manually. Messages contain repeat entries, missing session choices and later changes. Improve the process before considering automation.
A workflow with a single source of truth and a test table of expected outcomes.
Handle duplicates, missing information and changes. Use fictional records. Keep a manual fallback.
AI offers a large automated system before asking how sign-ups work.
What do you do first?
Write your choice and reason before reading the consequences below.
Practice: Write a five-step manual process and identify the single source of truth.
Your fictional test set includes a new sign-up, a repeated sign-up and a record without a session.
What should the test table contain?
Write your choice and reason before reading the consequences below.
Practice: Write three test cases with an expected outcome for each.
A participant changes sessions after a reminder has been sent.
Which update is safest for consistency?
Write your choice and reason before reading the consequences below.
Practice: Revise the workflow for a session change and add a fourth test case.
Two fictional participants share the same display name. Explain why name-only matching fails and propose a minimal distinguishing field for this exercise.
Avoid assuming names are unique. Use a fictional participant ID or another agreed identifier; collect only what the workflow needs.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 5 / Problem-solving apprentice
A fictional sheet reports weekly supply use. Some totals look plausible but rely on the wrong range or an unsupported interpretation. Learn to test the method rather than accept a formatted result.
A calculation note with a baseline, a corrected rule and independent test cases.
Reproduce calculations. Explain the starting value. Distinguish a formula result from a business claim.
Use rose from 20 units to 30 units. AI says the increase is 33.3%, using the new total as the denominator.
What should you check?
Write your choice and reason before reading the consequences below.
Practice: Explain the correction and calculate the percentage increase from 40 to 50.
A three-row fictional list contains 4, 6 and 10 units. A total displays 10 because the formula includes only the first two rows.
What is a good test?
Write your choice and reason before reading the consequences below.
Practice: Write the expected total and a test that adds a fourth row.
The corrected sheet shows higher supply use. AI concludes the team is inefficient. No production or attendance data is supplied.
What can you responsibly say?
Write your choice and reason before reading the consequences below.
Practice: Write a two-sentence summary and one follow-up question.
Use falls from 30 units back to 20. Calculate the percentage decrease and explain why it is not 50%.
The decrease is 10/30, about 33.3%. The comparison base changed. Explain the asymmetry without assuming the original calculation was wrong.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
MISSION 6 / Independent learner
You have practised facts, communication, constraints and tests. Now prepare a small learning artifact another person can use and defend your own decisions.
A project handoff with criteria, verification, an independent variation and a next-review plan.
Show your contribution. State limits. Test a fresh case without the model answer open.
You have an attractive AI-generated plan but cannot explain one of its steps.
What should you do before handing it off?
Write your choice and reason before reading the consequences below.
Practice: Choose one step from an earlier mission and explain it using a new small example.
A reviewer asks whether your skill transfers.
Which demonstration is strongest?
Write your choice and reason before reading the consequences below.
Practice: Select an earlier mission’s changed situation, answer it unaided and state one boundary of your conclusion.
The project is finished and you want to keep the skill.
What is a useful next step?
Write your choice and reason before reading the consequences below.
Practice: Write a review date, a new task, two success criteria and a fallback practice step.
Transfer one skill to your own work, study or hobby. State which parts of the fictional mission no longer apply and how you will test the adapted method.
Make the outcome observable, avoid claiming broad mastery from one task, and include a check that does not depend on copying the worked example.
Record your final artifact, your independent variation and what you can now explain. Note any support you used.
Choose a useful artifact from your own work, study or hobby. Record these nine parts:
Review purpose, accuracy, reasoning, transfer and ownership. The interactive tool can save this evidence and export a portfolio.