Try first, learn, practise, check and apply. This guide contains all lesson explanations, examples, practice briefs and decision answers. Use the interactive academy to save your work. Completion is self-review, not certification.
Use your browser’s Print command to save a PDF. If studying a decision question, attempt it before reading its answer.
1. Start with curiosity
Find your first useful task and keep control of the learning process.
1. What AI can do for you
Try first: Explain what you already think about this topic and what you want to find out.
AI systems can help recognize patterns, classify information or generate new content. Start with a small task whose result you can inspect. A useful assistant is not automatically a reliable authority.
Practise: List one everyday task, the help you want and how you will check the result.
Decision check: Which first task is easiest to evaluate?
Draft a short invitation using facts you supply
Let AI make every decision for an unfamiliar project
Accept an answer because it sounds professional
Answer: Draft a short invitation using facts you supply
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
2. AI is more than a chatbot
Try first: Explain what you already think about this topic and what you want to find out.
A chatbot is an interface for conversation. AI also appears in recommendations, image recognition and forecasting. Different systems have different inputs, outputs and limitations.
Practise: Compare two AI uses you encounter. Name their inputs, outputs and one possible error.
Decision check: What distinguishes an interface from a capability?
Every AI system is a conversational assistant
A friendly interface makes the underlying system accurate
Chat is a way to interact, not a complete description of the system
Answer: Chat is a way to interact, not a complete description of the system
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
3. Choose a human purpose
Try first: Explain what you already think about this topic and what you want to find out.
Begin with a need rather than a tool feature. Your goal determines whether AI adds value. Sometimes a checklist, calculator or conversation is simpler.
Practise: Describe a task where AI helps and one where a simpler tool is sufficient.
Decision check: You only need to total five known numbers. What is a sensible first choice?
Assume AI is required because it is available
Use a calculator and inspect the inputs
Use a long creative prompt to avoid arithmetic tools
Answer: Use a calculator and inspect the inputs
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
4. Turn curiosity into an outcome
Try first: Explain what you already think about this topic and what you want to find out.
An outcome says what you will do, under which conditions and how success will be checked. Time spent learning is an input, not evidence that you can perform.
Practise: Write a small goal with a performance, conditions and a check.
Decision check: Which is an observable learning outcome?
Produce a summary with every action traceable to the notes
Watch AI videos for an afternoon
Feel like an AI expert
Answer: Produce a summary with every action traceable to the notes
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
5. Try before asking
Try first: Explain what you already think about this topic and what you want to find out.
A baseline attempt reveals what you already know. Preserve it so you can compare your own reasoning with later work. You do not need a polished answer to start.
Practise: Make a two-minute unaided attempt at your goal. Record one difficulty without repairing it yet.
Decision check: Why save an early attempt?
It guarantees that later changes are improvements
It should be hidden because mistakes are embarrassing
It supplies evidence of your starting point
Answer: It supplies evidence of your starting point
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
6. The Vibe Learning loop
Try first: Explain what you already think about this topic and what you want to find out.
Use curiosity to choose a goal, try something, seek focused help, practise, check and reflect. The loop is flexible: evidence may send you back to a smaller gap.
Practise: Map your goal to try, learn, practise, check and apply. Include one place you may need to loop back.
Decision check: What should determine your next learning step?
The most impressive feature in the tool
The gap revealed by your attempt and check
The number of pages you have opened
Answer: The gap revealed by your attempt and check
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
7. Choose a manageable first tool
Try first: Explain what you already think about this topic and what you want to find out.
Compare tools against your task, access needs and data constraints. Features and account rules change, so inspect the current provider information before committing.
Practise: Create a three-criterion checklist for selecting a tool. Try the same fictional task if you have access.
Decision check: What is a useful selection method?
Compare tools on a real task and explicit criteria
Choose only by the longest feature list
Assume every free account has identical conditions
Answer: Compare tools on a real task and explicit criteria
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
8. Set up a safe practice space
Try first: Explain what you already think about this topic and what you want to find out.
Use fictional or public material while learning. Avoid putting passwords, private messages or identifiable records into an unfamiliar service. Inspect the service and your organization’s rules when real data is involved.
Practise: Rewrite a realistic practice scenario using entirely fictional details.
Decision check: What belongs in a beginner practice prompt?
A real password so the example is realistic
An entire private inbox without permission
Invented customer details that preserve the task
Answer: Invented customer details that preserve the task
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
9. Keep a learning notebook
Try first: Explain what you already think about this topic and what you want to find out.
Record the task, your attempt, assistance used, result and next question. A useful notebook preserves evidence rather than collecting attractive outputs with no context.
Practise: Create a notebook entry for your first experiment, including something that did not work.
Decision check: What makes a saved AI output useful for learning?
It is the longest answer in the conversation
Its task, checks and revisions are recorded with it
It looks polished without further explanation
Answer: Its task, checks and revisions are recorded with it
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
10. Project: your first useful AI task
Try first: Explain what you already think about this topic and what you want to find out.
Complete one low-stakes task from start to finish. Keep the original attempt, use a limited AI role and inspect the result. End by doing a small variation yourself.
Practise: Build a first-task portfolio: goal, baseline, prompt, result check and an unaided variation.
Decision check: What is the strongest project evidence?
An inspected result and a fresh attempt you can explain
A screenshot of an impressive response alone
A claim that the model did everything correctly
Answer: An inspected result and a fresh attempt you can explain
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
2. Understand the machinery
Build a practical mental model without needing a technical background.
11. Patterns rather than magic
Try first: Explain what you already think about this topic and what you want to find out.
Machine learning systems learn statistical patterns from data. Their performance depends on data, design and the conditions of use. Familiar-looking output can still be wrong.
Practise: Explain to a beginner why a system that works on familiar examples may fail on a new kind of input.
Decision check: What can change a model’s performance?
Whether its response uses confident wording
Whether the user wants it to be correct
A shift between its training examples and current inputs
Answer: A shift between its training examples and current inputs
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
12. Training and using a model
Try first: Explain what you already think about this topic and what you want to find out.
Training changes model parameters using data. Using a trained model to produce an output is inference. A chat message becoming part of the current conversation is not the same as immediately retraining the model.
Practise: Explain training and inference using a cooking or sports analogy, then name where the analogy breaks.
Decision check: Does one correction in a chat prove global retraining occurred?
Yes; if the assistant says it will remember
No; current context and model training are different processes
Yes; every correction immediately changes all users’ models
Answer: No; current context and model training are different processes
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
13. Tokens and text
Try first: Explain what you already think about this topic and what you want to find out.
Language models process text as tokens, which may be words, word parts or other units. Token counts are not identical to word counts and can differ by language or content.
Practise: Explain why a word count is an imperfect estimate of what fits in a model’s input.
Decision check: Which statement about tokens is accurate?
Tokens are processing units and need not be whole words
Every token is exactly one English word
All languages use identical tokens for equivalent sentences
Answer: Tokens are processing units and need not be whole words
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
14. Context has limits
Try first: Explain what you already think about this topic and what you want to find out.
A model can use only the context made available to it for the current operation. Long conversations may omit or compress earlier material. Restate essential constraints when accuracy matters.
Practise: Write a compact context card with your goal, audience, source facts and constraints.
Decision check: How can you reduce dependence on distant chat history?
Assume everything ever typed is always active
Add more unrelated history to make the chat larger
Restate essential information in a concise context card
Answer: Restate essential information in a concise context card
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
15. Generation is not a database lookup
Try first: Explain what you already think about this topic and what you want to find out.
A language model generates plausible sequences. Without an appropriate retrieval or verification step, a fluent answer may contain invented facts or sources.
Practise: Invent a harmless fictional claim and describe the evidence you would need before presenting it as fact.
Decision check: What does fluent writing establish?
That uncertainty has been eliminated
That the system produced fluent writing, not that the facts are verified
That a reliable source was consulted
Answer: That the system produced fluent writing, not that the facts are verified
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
16. Why answers vary
Try first: Explain what you already think about this topic and what you want to find out.
Generation can produce different responses to similar requests. Tool configuration and model versions can also affect outputs. Evaluate several cases instead of treating one lucky answer as a guarantee.
Practise: Design a three-case test for a summarization prompt.
Decision check: What is a better reliability check?
Test multiple relevant cases with the same criteria
Keep only the best-looking response
Assume a single success guarantees future results
Answer: Test multiple relevant cases with the same criteria
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
17. Capabilities are uneven
Try first: Explain what you already think about this topic and what you want to find out.
A system may be strong on one task and weak on another. Evaluate the specific activity, language and input type you need rather than relying on a general reputation.
Practise: List three tasks and define a different check for each.
Decision check: What should a tool evaluation focus on?
The confidence of its introduction
Its ability to describe itself as intelligent
Performance on the actual task and relevant cases
Answer: Performance on the actual task and relevant cases
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
18. Search and generation differ
Try first: Explain what you already think about this topic and what you want to find out.
Search retrieves indexed material; generation creates an output. Some products combine them, but you still need to distinguish a retrieved source from the model’s interpretation.
Practise: Describe a workflow that uses search to find a source and AI to help you interrogate it.
Decision check: A response includes links. What remains necessary?
Treat link count as a quality score
Open relevant sources and compare them with the claims
Assume every sentence comes directly from a source
Answer: Open relevant sources and compare them with the claims
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
19. Multimodal does not mean all-knowing
Try first: Explain what you already think about this topic and what you want to find out.
Some systems accept images, audio or other inputs alongside text. Their interpretations can miss detail or confuse context. The ability to accept a file is not proof of accurate analysis.
Practise: Plan a check for an AI description of a photograph containing small text.
Decision check: What should you do with uncertain extracted text?
Compare it with the original and mark unreadable parts
Answer: Compare it with the original and mark unreadable parts
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
20. Project: explain AI to a newcomer
Try first: Explain what you already think about this topic and what you want to find out.
A useful mental model covers patterns, context, generation and checking. Explain enough to support sensible use without claiming that the system understands exactly as a person does.
Practise: Create a one-page beginner explanation with an example, a limitation and a task-specific check.
Decision check: What improves a beginner explanation?
A human analogy presented as literally true
Technical terms without explanation
A concrete mechanism, a useful example and clear limits
Answer: A concrete mechanism, a useful example and clear limits
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
3. Have better AI conversations
Practise clear requests, context, constraints and iterative feedback.
21. Name the task
Try first: Explain what you already think about this topic and what you want to find out.
A prompt should make the requested action clear. Explain whether you want a summary, critique, list, draft or question. Vague requests leave the system to guess the job.
Practise: Write two prompts for the same text: one for summarizing and one for critique.
Decision check: Which request makes the action clear?
Make it better somehow
Identify two ambiguous instructions in this paragraph
Do something clever with this
Answer: Identify two ambiguous instructions in this paragraph
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
22. Describe the audience
Try first: Explain what you already think about this topic and what you want to find out.
Audience affects vocabulary, assumed knowledge and examples. Describe what the reader knows and what they need to do after reading. An audience label such as professional is less useful than an explicit description of the decision the reader needs to make.
Practise: Adapt a prompt for a beginner and for an experienced colleague. Explain what changes.
Decision check: Which audience description is actionable?
A first-time user who needs to restore a deleted file
Everyone everywhere
A very smart audience with no other context
Answer: A first-time user who needs to restore a deleted file
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
23. Provide the relevant context
Try first: Explain what you already think about this topic and what you want to find out.
Give information that changes the answer. Irrelevant detail can obscure the goal; missing constraints can lead to a plausible but unusable result. If removing a detail would not change a sensible answer, it may not belong in the prompt. Keep the source facts distinguishable from your preferences.
Practise: Write a brief context block for planning a fictional community event.
Decision check: Which context belongs in an event-plan prompt?
Unrelated personal messages
A long biography with no planning relevance
Budget, attendance, location constraints and objective
Answer: Budget, attendance, location constraints and objective
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
24. Specify useful output
Try first: Explain what you already think about this topic and what you want to find out.
An output format should serve a purpose. Ask for a checklist when you need actions, a table for comparisons or a short paragraph for a message. Formatting does not ensure factual accuracy.
Practise: Design an output format for turning fictional meeting notes into actions.
Decision check: What should the model do when a due date is absent?
Omit all actions with incomplete information without explanation
Mark it unknown rather than invent one
Choose a plausible deadline silently
Answer: Mark it unknown rather than invent one
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
25. Set constraints that matter
Try first: Explain what you already think about this topic and what you want to find out.
Constraints express boundaries such as length, available resources or exclusions. Too many conflicting constraints can make a task impossible. Prioritize the ones that affect success.
Practise: Write a prompt with two essential constraints and one optional preference.
Decision check: What should you do with conflicting constraints?
Resolve or prioritize them explicitly
Assume the model can satisfy an impossible combination
Repeat both constraints more forcefully
Answer: Resolve or prioritize them explicitly
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
26. Use examples deliberately
Try first: Explain what you already think about this topic and what you want to find out.
An example can demonstrate format, tone or the standard you want. Explain what to copy and what should change. Avoid examples that accidentally introduce wrong facts.
Practise: Create a sample output and label its reusable structure versus its fictional details.
Decision check: What risk does an example introduce?
Examples always reduce accuracy
An example guarantees the next answer matches every requirement
The model may copy details that were only illustrative
Answer: The model may copy details that were only illustrative
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
27. Ask for clarification
Try first: Explain what you already think about this topic and what you want to find out.
When important information is missing, invite a question before a full answer. Limit questions to what changes the result so the conversation stays useful.
Practise: Write a prompt that asks at most two necessary clarification questions.
Decision check: When is clarification most useful?
Only after a final output has been published
When missing information could materially change the answer
Before every trivial action regardless of context
Answer: When missing information could materially change the answer
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
28. Iterate one variable at a time
Try first: Explain what you already think about this topic and what you want to find out.
Change a specific element and inspect its effect. If you change task, context and format together, it becomes harder to know what helped. Use the same input and judging criteria for both versions. Record a tradeoff if one version improves clarity but loses an important detail.
Practise: Design an A/B prompt experiment with one changed instruction and fixed source material.
Decision check: What makes a prompt comparison interpretable?
A clear change and shared evaluation criteria
Different tasks and different criteria for each version
Selecting whichever answer is longer
Answer: A clear change and shared evaluation criteria
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
29. Ask for an inspectable rationale
Try first: Explain what you already think about this topic and what you want to find out.
Request a concise explanation, assumptions, sources or calculation steps you can check. A polished rationale is still an output to evaluate, not a guarantee about internal reasoning.
Practise: Write a prompt asking for a recommendation with assumptions and a testable justification.
Decision check: Which request supports evaluation?
Reveal a perfect hidden thought process
Guarantee your answer by saying you reasoned carefully
State the assumptions and evidence supporting your recommendation
Answer: State the assumptions and evidence supporting your recommendation
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
30. Project: your reusable prompt
Try first: Explain what you already think about this topic and what you want to find out.
A reusable prompt separates fixed instructions from changing inputs. Include the task, context slots, output format and checks. Test it on more than one example.
Practise: Build a prompt template and test it with two fictional inputs. Record one failure and revision.
Decision check: What makes a template ready to reuse?
It contains many impressive role instructions
It has been checked on relevant examples and its limits are recorded
It worked once on its original example
Answer: It has been checked on relevant examples and its limits are recorded
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
4. Learn with AI without losing the learning
Use coaching, recall, examples and feedback to strengthen your own performance.
31. Ask for coaching, not replacement
Try first: Explain what you already think about this topic and what you want to find out.
Coaching preserves decisions for the learner. Ask for a question or limited hint before a complete answer. You still need to attempt and check the work.
Practise: Write a coaching prompt for a skill you want to improve, specifying what the AI should not do yet.
Decision check: Which instruction keeps the learner active?
Ask one question and wait for my attempt
Complete every step before I try
Give me only the final answer to memorize
Answer: Ask one question and wait for my attempt
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
32. Break a skill into parts
Try first: Explain what you already think about this topic and what you want to find out.
A broad skill contains smaller performances. Identify a bottleneck you can practise and check. Avoid spending all your time on a comfortable subskill. Recheck the whole task after practising the part. Improvement in an isolated exercise may not transfer automatically to the complete performance.
Practise: Break a personal goal into four subskills and pick one using evidence from an attempt.
Decision check: How should you choose a practice target?
Always choose the easiest part
Choose the part with the most impressive terminology
Use a gap that limits the desired performance
Answer: Use a gap that limits the desired performance
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
33. Learn from a worked example
Try first: Explain what you already think about this topic and what you want to find out.
A worked example makes decisions visible. Before copying it, explain why each important step is present. Then change part of the problem and attempt it yourself.
Practise: Annotate a short worked example with the purpose of three steps.
Decision check: What turns an example into active learning?
Saving it without returning to it
Explaining the decisions and attempting a variation
Copying its appearance without inspecting the steps
Answer: Explaining the decisions and attempting a variation
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
34. Fade the support
Try first: Explain what you already think about this topic and what you want to find out.
Support can decrease as you gain control: full example, partial hints, then independent performance. Use evidence to decide when to remove help or restore a small hint.
Practise: Design three levels of support for one practice task.
Decision check: What is the purpose of fading support?
To test and build increasing independence
To make every task harder regardless of readiness
To remove all help before the learner starts
Answer: To test and build increasing independence
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
35. Retrieve before reviewing
Try first: Explain what you already think about this topic and what you want to find out.
Try to produce the idea without looking. Compare afterward and correct a specific gap. Familiarity while reading is not the same as being able to explain or use the idea.
Practise: Write three recall questions for a topic, attempt them and note one gap.
Decision check: What does a closed-note attempt reveal?
Whether the explanation is attractive
A permanent limit on your ability
What you can produce without the visible explanation
Answer: What you can produce without the visible explanation
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
36. Explain it in your own words
Try first: Explain what you already think about this topic and what you want to find out.
A useful explanation connects a mechanism, an example and a boundary. Rephrasing words without understanding the relationship can leave the same gap hidden.
Practise: Explain one AI concept to a beginner using an example and a limitation.
Decision check: Which explanation is most informative?
One that repeats the definition faster
One that shows how the idea works and where it stops applying
One that swaps every word for a synonym
Answer: One that shows how the idea works and where it stops applying
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
37. Use feedback as a next action
Try first: Explain what you already think about this topic and what you want to find out.
Ask feedback to identify a specific gap against a criterion. Convert it into a revision or practice task. Generic praise is not evidence that the result works.
Practise: Write a feedback request and a follow-up task that addresses one likely weakness.
Decision check: Which feedback helps you practise?
The deadline is missing; add it and ask a reader to identify it
Excellent work, keep going
This feels wrong, without further detail
Answer: The deadline is missing; add it and ask a reader to identify it
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
38. Space your return visits
Try first: Explain what you already think about this topic and what you want to find out.
Plan later attempts rather than relying only on immediate repetition. Choose intervals that fit your goal and adjust based on what you can retrieve. A saved date is a plan, not a guarantee of retention.
Practise: Create a two-visit review plan with a recall task and a new application.
Decision check: What should a review visit include?
Only rereading the highlighted passage
A count of how many times the page opened
An attempt before looking and a check afterward
Answer: An attempt before looking and a check afterward
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
39. Mix related challenges
Try first: Explain what you already think about this topic and what you want to find out.
Practising related task types together can require choosing an approach rather than repeating a known routine. First understand the individual tasks, then compare when each applies.
Practise: Create three short situations that require different AI assistance modes.
Decision check: What does mixed practice add?
A reason to skip learning the individual skills
A decision about which approach fits the task
A guarantee that all practice becomes easier
Answer: A decision about which approach fits the task
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
40. Project: your personal AI coach
Try first: Explain what you already think about this topic and what you want to find out.
Design a coaching routine that includes an initial attempt, limited help, feedback and an unaided transfer task. Decide what evidence will show improvement.
Practise: Run a coaching session or simulate it using the lesson examples. Preserve the before and after work.
Decision check: What belongs in a coaching portfolio?
Your attempts, assistance used and an independent check
Only the coach’s final answer
A claim of mastery based on time spent
Answer: Your attempts, assistance used and an independent check
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
5. Verify and think critically
Check claims, calculations, sources and uncertainty before relying on outputs.
41. Separate claims from wording
Try first: Explain what you already think about this topic and what you want to find out.
A confident style can make a weak claim feel strong. Extract the factual statements and decide which require verification before use. Distinguish descriptive claims from opinions and recommendations. Each kind of statement may need a different check.
Practise: Take a fictional AI recommendation and list three claims that need evidence.
Decision check: What should you assess first?
How often the answer uses the word definitely
Whether the answer flatters your idea
The checkable claims, not the confident tone
Answer: The checkable claims, not the confident tone
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
42. Open the source
Try first: Explain what you already think about this topic and what you want to find out.
A citation may be missing, invented or irrelevant. Open the source and check whether it supports the exact statement, including scope and conditions. A source can be genuine but still fail to support the conclusion. Record the passage or evidence that actually bears on the claim.
Practise: Write a source-check checklist for a claim you might include in a report.
Decision check: What does a citation alone establish?
The source was read accurately by the model
A reference was presented; support still needs checking
The claim is necessarily correct
Answer: A reference was presented; support still needs checking
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
43. Check dates and scope
Try first: Explain what you already think about this topic and what you want to find out.
Information can be accurate for one time, place or product version and wrong for another. Match the claim to the conditions in which you will use it. Preserve qualifiers such as trial participants or a named plan. Removing them can change a narrowly correct statement into a misleading generalization.
Practise: Identify the date, population or version limits you would check for an AI tool recommendation.
Decision check: Which detail can invalidate an otherwise accurate answer?
It describes a different product version or population
It uses short sentences
It includes a table
Answer: It describes a different product version or population
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
44. Recalculate a number
Try first: Explain what you already think about this topic and what you want to find out.
Use an independent method to check arithmetic and units. A correct-looking explanation can contain a numerical error. Small test cases make mistakes easier to find.
Practise: Create a small fictional budget, calculate its total and explain how you would audit an AI result.
Decision check: What is a strong numerical check?
Ask the same model whether it feels certain
Accept a total because it has two decimal places
Recompute from the original inputs and units
Answer: Recompute from the original inputs and units
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
45. Distinguish correlation and cause
Try first: Explain what you already think about this topic and what you want to find out.
Two things changing together does not by itself establish that one caused the other. Consider alternative explanations and what a comparison could reveal.
Practise: List two alternative explanations for an apparent improvement after adopting AI.
Decision check: What can an uncontrolled before-and-after comparison support?
Proof that all other influences were absent
An observed change, with causal uncertainty
Proof that AI caused the whole change
Answer: An observed change, with causal uncertainty
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
46. Compare genuinely different evidence
Try first: Explain what you already think about this topic and what you want to find out.
Multiple answers may repeat the same underlying source or error. Seek relevant evidence with different origins and inspect how each supports the claim.
Practise: Design a comparison that distinguishes three copied claims from independent supporting evidence.
Decision check: What makes corroboration stronger?
Relevant evidence with distinct origins and methods
Repeating the same claim in three chat windows
Counting links without opening them
Answer: Relevant evidence with distinct origins and methods
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
47. Make uncertainty explicit
Try first: Explain what you already think about this topic and what you want to find out.
Separate what is supported, what is inferred and what remains unknown. This helps a reader decide how to use the result and what to check next. A useful unknown is specific: name the missing fact and how it would affect the decision. Vague disclaimers are less actionable.
Practise: Rewrite a confident fictional recommendation to mark one fact, one inference and one unknown.
Decision check: How should missing information be handled?
Fill it with a plausible detail silently
Hide it so the answer appears complete
Label it and identify a useful next check
Answer: Label it and identify a useful next check
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
48. Test the edge case
Try first: Explain what you already think about this topic and what you want to find out.
A solution may work on a typical input but fail when information is missing, contradictory or unusual. Choose edge cases that matter for your intended use.
Practise: Design three edge cases for an AI task you use.
Decision check: What should a summary do when there are no decisions?
Copy a decision from an earlier example
State that no decisions are recorded
Invent a likely decision to satisfy the template
Answer: State that no decisions are recorded
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
49. Know when to stop and ask
Try first: Explain what you already think about this topic and what you want to find out.
Some tasks need expertise, authority or verification you do not have. Recognize when an AI output should remain a draft or a question for a qualified person.
Practise: Name a task where you would use AI for preparation but seek an authorized human decision.
Decision check: What is a sensible role for AI in an uncertain consequential task?
Help organize questions and evidence for a qualified reviewer
Replace required expertise because it answers quickly
Conceal uncertainty so a decision can happen sooner
Answer: Help organize questions and evidence for a qualified reviewer
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
50. Project: an evidence audit
Try first: Explain what you already think about this topic and what you want to find out.
Build a short audit with a claim, source, check, uncertainty and decision. A useful audit may conclude that a claim should not be used. Keep planned checks separate from checks you performed. If a source cannot be accessed, say so instead of treating its title as confirmation.
Practise: Audit three claims from a real low-stakes answer or a fictional example. Mark verification you actually performed.
Decision check: What makes an audit trustworthy?
Every claim receives a positive conclusion
Uncertainty is removed from the final record
It distinguishes completed checks from planned checks
Answer: It distinguishes completed checks from planned checks
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
6. Use AI responsibly
Protect people, information and trust while making practical choices.
51. Minimize the data you share
Try first: Explain what you already think about this topic and what you want to find out.
Provide only the information needed for the task. Removing a name may not be enough if other details identify someone. Use fictional examples when possible.
Practise: Reduce a fictional customer record to the minimum details needed for a draft reply.
Decision check: What is a good default for practice data?
Assume removing a first name guarantees anonymity
Use fictional details and omit unnecessary information
Include every available field just in case
Answer: Use fictional details and omit unnecessary information
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
52. Inspect data settings
Try first: Explain what you already think about this topic and what you want to find out.
Providers may offer different retention, training and sharing controls. Read the current settings and applicable agreements rather than assuming all tools behave the same.
Practise: Create a checklist of settings and organizational rules to inspect before using real work material.
Decision check: What determines whether real work data can be used?
Current service terms, settings and relevant organizational rules
The interface color
A friend’s experience with a different account
Answer: Current service terms, settings and relevant organizational rules
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
53. Keep credentials out of prompts
Try first: Explain what you already think about this topic and what you want to find out.
Passwords, API keys and recovery codes are secrets, not useful context for most learning tasks. Use placeholders and keep credentials in appropriate secure systems.
Practise: Redact a fictional setup example and explain which values must never be shared.
Decision check: What should replace a real credential in a learning example?
A screenshot of the full secret
The secret split across two messages
A clearly marked placeholder
Answer: A clearly marked placeholder
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
54. Notice unfair assumptions
Try first: Explain what you already think about this topic and what you want to find out.
Outputs can reflect stereotypes or uneven representation. Examine who is described, omitted or judged differently. Evaluate the actual task rather than demographic shortcuts.
Practise: Review a fictional description for an unsupported assumption and rewrite it around relevant evidence.
Decision check: What should guide a comparison of people’s work?
A model’s unsupported personality guess
Relevant evidence and explicit criteria
Stereotypes that sound familiar
Answer: Relevant evidence and explicit criteria
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
55. Consider accessibility
Try first: Explain what you already think about this topic and what you want to find out.
An output should work for its audience. Offer clear structure, understandable language and appropriate alternatives to purely visual or audio information.
Practise: Adapt a short AI-generated instruction for a reader who needs plain language and clear steps.
Decision check: What improves access to a chart’s meaning?
A concise text explanation of the important information
Only changing its colors
Adding more visual decoration without explanation
Answer: A concise text explanation of the important information
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
56. Respect authorship and permission
Try first: Explain what you already think about this topic and what you want to find out.
Check whether you have permission to use supplied material and how attribution should work in your setting. AI assistance does not remove the need to respect other people’s work.
Practise: Write a disclosure statement for an AI-assisted personal project, identifying your own contribution.
Decision check: What remains your responsibility when using AI?
Assuming generated output has no usage constraints
Claiming all assisted work was created without help
Checking permissions, attribution expectations and the final work
Answer: Checking permissions, attribution expectations and the final work
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
57. Avoid impersonation
Try first: Explain what you already think about this topic and what you want to find out.
Synthetic text, images and audio can mislead people about who said or did something. Label fictional demonstrations and obtain appropriate permission when using someone’s likeness or voice.
Practise: Design a clearly labeled fictional AI demonstration that cannot be mistaken for a real endorsement.
Decision check: What preserves trust in a simulated testimonial?
Removing all context so it seems authentic
Labeling it fictional and avoiding a real person’s identity without permission
Making it look like a genuine customer quote
Answer: Labeling it fictional and avoiding a real person’s identity without permission
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
58. Recognize instructions inside data
Try first: Explain what you already think about this topic and what you want to find out.
A document or webpage can contain text that tries to redirect an AI tool. Treat outside content as material to inspect, not automatically as authorized instructions.
Practise: Describe how you would handle an instruction embedded in material you asked AI to summarize.
Decision check: What authority should a retrieved document have?
It supplies task data, not permission to override the user
It can authorize any action if written forcefully
It should control the tool whenever it mentions AI
Answer: It supplies task data, not permission to override the user
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
59. Keep humans in consequential decisions
Try first: Explain what you already think about this topic and what you want to find out.
As impact rises, review, accountability and appropriate expertise matter more. Decide what the system may suggest and what a person must approve. A reviewer needs enough time, context and authority to reject a proposal. A checkbox with no real inspection is not meaningful oversight.
Practise: Define an approval boundary for a fictional AI-assisted workplace workflow.
Decision check: What is a useful human review point?
Only after an irreversible action fails
Never, if the draft is fluent
Before a consequential action, with access to the evidence
Answer: Before a consequential action, with access to the evidence
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
60. Project: your responsible-use agreement
Try first: Explain what you already think about this topic and what you want to find out.
Make a short agreement covering data, checks, disclosure and action boundaries. Tie each rule to a real task rather than writing vague promises. Include a response for when a rule cannot be satisfied: use fictional data, request permission, seek review or stop the action.
Practise: Write a one-page agreement for your own AI practice and test it against two fictional scenarios.
Decision check: What makes a responsible-use rule practical?
It assumes the model handles every responsibility
It specifies what to do in a concrete situation
It only says use AI ethically
Answer: It specifies what to do in a concrete situation
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
7. Everyday productivity with judgment
Create useful drafts, summaries and plans while keeping facts and decisions visible.
61. Summarize without inventing
Try first: Explain what you already think about this topic and what you want to find out.
A summary should preserve the source’s important meaning while reducing length. Specify what matters and compare the result with the original, especially names, decisions and exceptions.
Practise: Summarize a short fictional note and mark every statement that needs source comparison.
Decision check: What should happen to a proposed date in a summary?
It remains clearly proposed
It becomes confirmed to simplify the writing
It is replaced by the most likely date
Answer: It remains clearly proposed
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
62. Extract action items
Try first: Explain what you already think about this topic and what you want to find out.
Extraction selects information that is present. Separate tasks, owners and deadlines, and mark missing fields. Do not silently turn ideas into commitments.
Practise: Create a fictional note containing an idea and an action, then extract only the commitment.
Decision check: Which statement is an assigned action?
Maybe we should have a speaker someday
A speaker might be interesting
Jo will contact the speaker by Thursday
Answer: Jo will contact the speaker by Thursday
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
63. Draft a clear email
Try first: Explain what you already think about this topic and what you want to find out.
Provide the purpose, audience and verified facts. Inspect whether the draft makes the requested action clear and preserves your intended tone. Check names, attachments and deadlines yourself. A confident draft may accidentally add a promise you did not intend to make.
Practise: Draft a reminder for a fictional event, then remove unsupported details and unnecessary sentences.
Decision check: What is the most important check before sending?
The model says the tone is perfect
The request and facts match your intention
The email sounds more elaborate than your normal writing
Answer: The request and facts match your intention
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
64. Plan with constraints
Try first: Explain what you already think about this topic and what you want to find out.
A useful plan respects time, resources and dependencies. Ask the system to state assumptions and identify what must be confirmed before execution. Separate tasks that can happen in parallel from those that depend on an earlier decision. Add a fallback for an uncertain resource.
Practise: Plan a fictional one-hour workshop with limited materials and identify two dependencies.
Decision check: What makes a plan usable?
It fits the constraints and exposes dependencies
It includes every possible activity
It ignores setup time to look efficient
Answer: It fits the constraints and exposes dependencies
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
65. Brainstorm then choose
Try first: Explain what you already think about this topic and what you want to find out.
Idea generation and evaluation are different tasks. First explore options, then choose using explicit criteria. Do not confuse novelty with usefulness.
Practise: Create three approaches to a small goal and justify one using your criteria.
Decision check: How should you choose among generated ideas?
Pick the first idea because it arrived fastest
Pick the most surprising idea regardless of feasibility
Use criteria tied to the actual goal
Answer: Use criteria tied to the actual goal
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
66. Compare options fairly
Try first: Explain what you already think about this topic and what you want to find out.
Use the same criteria and comparable information for every option. Mark unknowns rather than filling gaps to make a neat table. Explain which criteria matter most to the decision. Equal-looking columns can conceal a difference that is essential for the user.
Practise: Build a fictional comparison table with two options and one unknown field.
Decision check: What makes a comparison misleading?
Using the same criteria for each option
Using different conditions without making them visible
Showing an unknown value explicitly
Answer: Using different conditions without making them visible
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
67. Edit while preserving meaning
Try first: Explain what you already think about this topic and what you want to find out.
Editing can improve clarity but also change intent. Ask for specific changes and compare important claims against the original. Pay special attention to quantities, qualifications and promises. Keep a version of the original so you can compare the change.
Practise: Write a cautious statement, revise its clarity and check that the level of commitment stayed the same.
Decision check: What must survive an edit?
The intended meaning and degree of certainty
Every original word regardless of clarity
The model’s preferred interpretation
Answer: The intended meaning and degree of certainty
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
68. Translate with context
Try first: Explain what you already think about this topic and what you want to find out.
Translation depends on audience, purpose and ambiguity. For important communication, use a competent reviewer and preserve names, numbers and specialized terms carefully.
Practise: Prepare a translation brief with audience, terminology and fields that must remain unchanged.
Decision check: What should you check in a translated announcement?
Only whether the text is longer
Whether the model claims native fluency
Meaning, names, dates and audience-appropriate language
Answer: Meaning, names, dates and audience-appropriate language
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
69. Build a reusable checklist
Try first: Explain what you already think about this topic and what you want to find out.
A checklist helps prevent omissions in repeatable work. Keep items observable and test it on a real case. Too many vague items can make it hard to use.
Practise: Create a five-item checklist for a recurring task and test it on a fictional output.
Decision check: Which checklist item is observable?
Ensure everything feels professional
Every action has an owner or an explicit unknown marker
Make it excellent
Answer: Every action has an owner or an explicit unknown marker
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
70. Project: your useful workflow
Try first: Explain what you already think about this topic and what you want to find out.
Combine preparation, generation, checking and revision into a repeatable process. Record where human judgment is required and what should never happen automatically.
Practise: Build and test a workflow on two fictional cases, including one with missing information.
Decision check: What should a workflow test include?
A normal case and a relevant difficult case
Only the easiest example
Only a description of the intended process
Answer: A normal case and a relevant difficult case
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
8. Create, analyse and build
Explore practical creative and technical tasks with explicit checks.
71. Write a creative brief
Try first: Explain what you already think about this topic and what you want to find out.
A brief names the audience, purpose, constraints and desired effect. It gives creative work direction without dictating every detail. Evaluate the result against the brief.
Practise: Write a creative brief for a fictional event, including two required facts and a visual constraint.
Decision check: What should guide creative evaluation?
Only whether the result looks surprising
The amount of detail in the prompt
The audience and purpose in the brief
Answer: The audience and purpose in the brief
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
72. Describe an image precisely
Try first: Explain what you already think about this topic and what you want to find out.
An image prompt can specify subject, composition, lighting and style. Generated images may distort text or details, so inspect what matters to the intended use.
Practise: Draft an image prompt and list the details you would inspect before using the result.
Decision check: Which generated detail needs checking?
Nothing if the image is visually appealing
Any text or factual feature important to the message
Only the overall color palette
Answer: Any text or factual feature important to the message
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
73. Describe visuals accessibly
Try first: Explain what you already think about this topic and what you want to find out.
Useful alternative text conveys the information or function of an image in its context. Decorative detail should not obscure the important meaning. The same image may need different descriptions in different contexts. Describe what a reader needs to understand or do here.
Practise: Write an accessible description of a simple fictional chart with three values.
Decision check: What should a chart description prioritize?
The information the chart communicates
Every decorative pixel
An enthusiastic judgment of its beauty
Answer: The information the chart communicates
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
74. Turn audio into usable notes
Try first: Explain what you already think about this topic and what you want to find out.
A transcript may mishear names, numbers or overlapping speech. Review important passages against the recording when permitted and keep uncertainty visible.
Practise: Design a transcript review checklist for a fictional workshop recording.
Decision check: How should an unclear number be handled?
Choose the more plausible number without noting it
Delete the surrounding context
Check the recording or mark it uncertain
Answer: Check the recording or mark it uncertain
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
75. Explore a small dataset
Try first: Explain what you already think about this topic and what you want to find out.
Start with a question, inspect the data and identify missing or inconsistent values. A model’s interpretation should be checked against the actual records.
Practise: Create a five-row fictional dataset with one missing value and describe how you would analyse it.
Decision check: What is a safe assumption about a blank cell?
It can be replaced with any convenient value
Its meaning needs checking before treating it as zero
It always means zero
Answer: Its meaning needs checking before treating it as zero
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
76. Choose an honest chart
Try first: Explain what you already think about this topic and what you want to find out.
Select a chart that supports the comparison and label units clearly. Avoid visual choices that exaggerate differences or imply an order that does not exist.
Practise: Choose a chart for three fictional categories and explain its scale and labels.
Decision check: What supports honest interpretation?
A suitable chart, clear labels and a defensible scale
Removing axis labels for a cleaner look
Choosing a scale only to maximize drama
Answer: A suitable chart, clear labels and a defensible scale
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
77. Read code before running it
Try first: Explain what you already think about this topic and what you want to find out.
Generated code is a proposal to inspect. Understand inputs, outputs and side effects before execution. Avoid running unfamiliar commands against important files or credentials.
Practise: Describe a safe inspection and test plan for a fictional file-renaming script.
Decision check: What should happen before running unfamiliar code?
Run it on the only copy of important files
Assume generated code cannot delete anything
Inspect its effects and test on disposable data
Answer: Inspect its effects and test on disposable data
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
78. Test a small program
Try first: Explain what you already think about this topic and what you want to find out.
Specify expected behavior and use small examples with known results. Include an edge case. A program running without an error is not proof that it returns the right answer.
Practise: Write three test cases for a function that totals a list of prices.
Decision check: What is a meaningful test?
Asking the model to praise its own code
Known inputs with explicit expected results
Checking that the code looks sophisticated
Answer: Known inputs with explicit expected results
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
79. Debug with evidence
Try first: Explain what you already think about this topic and what you want to find out.
Reproduce the problem, compare expected and actual behavior and test one hypothesis. Preserve a working copy so experiments are reversible. Record the expected result before trying a fix. After a repair, test a different relevant case to see whether you corrected the cause.
Practise: Write a debugging log for a fictional wrong total, including a rejected hypothesis.
Decision check: What makes debugging informative?
A reproducible case and a tested hypothesis
Changing several unrelated things at once
Replacing the output manually without finding the cause
Answer: A reproducible case and a tested hypothesis
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
80. Project: make a checked artifact
Try first: Explain what you already think about this topic and what you want to find out.
Build a small artifact such as a poster brief, data summary or code specification. Define acceptance criteria before making it and show evidence for each check.
Practise: Create one artifact with at least three acceptance criteria and a short record of verification.
Decision check: What makes an artifact portfolio useful?
Only the most attractive screenshot is shown
All limitations are omitted
The output is linked to criteria and checks
Answer: The output is linked to criteria and checks
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
9. Design reliable AI workflows
Move from isolated prompts to measured, reviewable processes.
81. Map the whole workflow
Try first: Explain what you already think about this topic and what you want to find out.
Identify inputs, transformations, checks and outputs before automating. A good prompt cannot repair every problem caused by missing data or unclear responsibility.
Practise: Draw or describe a five-step AI-assisted process with a named review point.
Decision check: What should a workflow map expose?
Only the tool’s branding
Data flow, decisions and checks
Only the generation step
Answer: Data flow, decisions and checks
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
82. Separate data from instructions
Try first: Explain what you already think about this topic and what you want to find out.
Clearly label source material and the task to perform on it. Labels help communication but do not by themselves eliminate malicious instructions or model mistakes.
Practise: Write a prompt with separate task, source and output sections.
Decision check: What does clear separation achieve?
It clarifies the task, while checks are still needed
It guarantees protection against every injection
It makes source content automatically trustworthy
Answer: It clarifies the task, while checks are still needed
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
83. Understand retrieval support
Try first: Explain what you already think about this topic and what you want to find out.
Retrieval can supply relevant documents to a model. The model may still retrieve the wrong passage or interpret it incorrectly. Check both source selection and answer support.
Practise: Design a retrieval check that asks whether the passage is relevant and whether the answer follows from it.
Decision check: Does retrieval guarantee a correct answer?
Yes; any retrieved page makes the answer correct
Yes; if the source list is long
No; relevance and interpretation still need checking
Answer: No; relevance and interpretation still need checking
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
84. Use structured output carefully
Try first: Explain what you already think about this topic and what you want to find out.
Structured formats make information easier to process, but valid structure does not guarantee correct content. Check required fields, types and meaning.
Practise: Specify fields for an action item and identify one structural and one factual check.
Decision check: What must be checked beyond valid JSON?
Whether the model likes the format
Whether the values are supported and appropriate
Only whether braces are balanced
Answer: Whether the values are supported and appropriate
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
85. Know what an API does
Try first: Explain what you already think about this topic and what you want to find out.
An API lets software request a service through a defined interface. Using one involves inputs, outputs, authentication and possible cost or failure. A browser chat and an API may have different terms and capabilities.
Practise: Describe an API request and response in plain language, including where a failure could occur.
Decision check: What should an application plan for?
Errors, access controls and validation of returned content
Only the happy path
An assumption that every request succeeds
Answer: Errors, access controls and validation of returned content
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
86. Limit an agent’s authority
Try first: Explain what you already think about this topic and what you want to find out.
An agentic workflow can choose steps or use tools. Define what it may read, change or send, and where approval is required. More autonomy increases the importance of boundaries.
Practise: Design permissions for an assistant that organizes a fictional project.
Decision check: Which permission is appropriately bounded?
Allow every action because the assistant is helpful
Hide tool actions from the user
Draft proposed actions and request approval before external changes
Answer: Draft proposed actions and request approval before external changes
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
87. Evaluate with a small test set
Try first: Explain what you already think about this topic and what you want to find out.
Create representative cases and explicit criteria before comparing prompts or tools. Include typical, missing-information and difficult cases. Keep test inputs consistent across alternatives.
Practise: Build a five-case evaluation plan for a task you use.
Decision check: What makes a comparison fair?
A decision based on one impressive answer
The same cases and criteria for each alternative
Different easy cases for your preferred option
Answer: The same cases and criteria for each alternative
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
88. Track failures as categories
Try first: Explain what you already think about this topic and what you want to find out.
Group failures so you can target improvements: missing context, unsupported claim, wrong format or incorrect reasoning. A category should suggest an action to test.
Practise: Create a failure log with category, example, proposed fix and retest result.
Decision check: Why categorize errors?
To choose and evaluate targeted changes
To hide inconvenient examples
To prove the user caused every failure
Answer: To choose and evaluate targeted changes
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
89. Balance effort and benefit
Try first: Explain what you already think about this topic and what you want to find out.
Consider setup, review time, cost and output quality together. A fast draft may require extensive repair. Measure the whole task rather than only generation speed.
Practise: Compare a manual and AI-assisted process using fictional times and explicit quality checks.
Decision check: What should a productivity comparison include?
Only seconds until the first draft appears
Only the length of the final output
Preparation, generation, checking and repair time
Answer: Preparation, generation, checking and repair time
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
90. Project: a workflow with a test plan
Try first: Explain what you already think about this topic and what you want to find out.
Create a bounded workflow, a small test set and a failure response. Record what works, what remains uncertain and when a person takes over. Define a failure response before deployment: retry, ask for missing information, send to a reviewer or decline to act. Choose the response according to impact.
Practise: Document and test your workflow on three cases, then write a decision about whether it is ready for your intended use.
Decision check: What supports a readiness decision?
The assistant’s assurance that it is production ready
Evidence from relevant tests and known limits
The number of workflow steps
Answer: Evidence from relevant tests and known limits
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
10. Transfer, teach and keep growing
Finish with an independent project and a sustainable learning practice.
91. Choose a meaningful capstone
Try first: Explain what you already think about this topic and what you want to find out.
Select a real but manageable problem with an audience and a checkable outcome. Avoid a project so broad that you cannot inspect the result. Decide what is outside scope so the project remains finishable. A small checked result is easier to learn from than an ambitious untested promise.
Practise: Write a capstone brief with audience, deliverable, constraints and success criteria.
Decision check: Which capstone is feasible to assess?
A small briefing checked against supplied notes and reader needs
Transform every part of my life with AI
Build the best system for everyone without criteria
Answer: A small briefing checked against supplied notes and reader needs
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
92. Plan evidence before production
Try first: Explain what you already think about this topic and what you want to find out.
Decide what you will save to show the process: baseline, prompts, outputs, checks and independent work. Evidence is easier to collect when planned in advance.
Practise: Create an evidence checklist for your capstone before making its final artifact.
Decision check: What should a portfolio demonstrate?
Only that you accessed an AI tool
Only the final aesthetic appearance
What you did, checked and can explain
Answer: What you did, checked and can explain
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
93. Work with a human reviewer
Try first: Explain what you already think about this topic and what you want to find out.
Choose a reviewer who can assess relevant criteria. Give a focused question rather than asking for general approval. Record disagreements and decide what to test.
Practise: Write a reviewer brief with two concrete questions and a way to record feedback.
Decision check: What is a useful reviewer question?
Can you say this is excellent?
Which action and deadline do you understand from this draft?
Do you like AI?
Answer: Which action and deadline do you understand from this draft?
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
94. Adapt to an unfamiliar context
Try first: Explain what you already think about this topic and what you want to find out.
Transfer requires deciding what remains the same and what changes. Keep the principle but adjust the example, audience or constraints. Test the adaptation independently.
Practise: Apply a familiar method to a new setting and explain one adaptation.
Decision check: What shows transfer?
Choosing how to apply a principle in a changed situation
Repeating the identical example from memory
Copying a template without inspecting its assumptions
Answer: Choosing how to apply a principle in a changed situation
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
95. Teach someone else
Try first: Explain what you already think about this topic and what you want to find out.
Teaching exposes gaps in your explanation. Ask the learner to attempt a new example and use their response to revise your instruction. Understanding is not measured by agreement alone.
Practise: Prepare a five-minute mini-lesson with an example, a misconception and a new task.
Decision check: What checks whether your teaching helped?
Their polite nod alone
The length of your explanation
A learner’s performance on a new relevant task
Answer: A learner’s performance on a new relevant task
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
96. Recover from a failed attempt
Try first: Explain what you already think about this topic and what you want to find out.
A failed task can reveal a specific gap. Separate errors in knowledge, process, tool choice or missing information. Choose a small repair and retest rather than abandoning the whole goal.
Practise: Write a recovery plan for one capstone failure, including an alternative explanation.
Decision check: What should follow a failed check?
Deleting the failure from the portfolio
A diagnosis, targeted repair and new check
A claim that the tool is always useless
Answer: A diagnosis, targeted repair and new check
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
97. Calibrate confidence with evidence
Try first: Explain what you already think about this topic and what you want to find out.
Compare how certain you felt with what you could actually demonstrate. Confidence is useful reflection, but it does not replace checks. Record both successes and unsupported assumptions.
Practise: Describe a moment when a check changed your confidence and what you will do differently.
Decision check: What should raise justified confidence?
Repeated relevant evidence with known limits
A more enthusiastic model response
Avoiding tests that could reveal mistakes
Answer: Repeated relevant evidence with known limits
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
98. Keep your skills current
Try first: Explain what you already think about this topic and what you want to find out.
Tools change, so maintain a habit of checking current documentation and retesting important workflows. Separate stable principles from product-specific instructions.
Practise: Identify two stable skills and two product details you will recheck periodically.
Decision check: What needs periodic verification?
Whether checking evidence matters at all
Whether a polished answer can ever be wrong
Current features, terms and behavior of important workflows
Answer: Current features, terms and behavior of important workflows
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
99. Create your next learning plan
Try first: Explain what you already think about this topic and what you want to find out.
Use your portfolio to choose the next gap, not just the next fashionable topic. Schedule a fresh attempt and define evidence that would show progress. Include an opportunity to apply the skill without the original example. Use the outcome to revise the plan rather than treating the schedule as fixed.
Practise: Write a two-week plan with one target skill, two practice sessions and an independent check.
Decision check: What should drive your next plan?
The desire to collect more completion marks
Evidence of a useful remaining gap
Only the newest tool announcement
Answer: Evidence of a useful remaining gap
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
100. Final project: demonstrate independent judgment
Try first: Explain what you already think about this topic and what you want to find out.
Present a useful artifact and explain the decisions behind it. Include AI assistance, checks, limitations and a fresh task completed with reduced help. Completion is a record of work, not a universal AI certification.
Practise: Submit your capstone in the portfolio: artifact description, evidence, disclosure, limitations and next challenge. Ask a relevant person to review it if possible.
Decision check: What is the strongest final demonstration?
A useful checked result plus reasoning and an independent variation
One flawless-looking AI answer with no process record
A claim that finishing 100 lessons proves expertise in every AI task
Answer: A useful checked result plus reasoning and an independent variation
Apply: Change the audience, inputs or constraints. Attempt a new situation without the worked example. Record your result, assistance, evidence and uncertainty.
Reflect: What changed in your understanding, and what will you test next?
Scope and further reading
Original practical curriculum reviewed 1 November 2026. Product features and account rules change. Check current official documentation. Further reading: UNESCO guidance and NIST Generative AI Profile.