Technology usage — 9 action sessions

Delegate carefully: map the work, prompt with discipline, verify outputs, protect data and automate safely. Each session produces a real artifact in about an hour.

About these sessions

Use these sessions to delegate carefully, supervise outputs, protect data, and turn tools into disciplined work support.

Each session includes a scenario, title concept definitions, quick self-check, action steps, worksheet, choice path, prompts, checkpoint, small project, evidence to save, mistakes to avoid, and finish line.

⚠️ Safety note. Learners should use prompts only with information they are allowed to share. For legal, financial, medical, psychological, cybersecurity, or compliance decisions, they should consult qualified professionals and approved organizational policies.

Pick a session in the menu — one session is shown at a time. Facilitating a group? Use the facilitator guide.

Session 1 — Map the Work Before Choosing the Tool

⏱ 55 minutes · 🎯 You will build: a workflow map that shows where technology can help and where human control must remain.

Start here

You are preparing a recurring activity such as a monthly report, client follow-up, purchase request, field visit, or training session. You feel tempted to ask an AI tool to handle it, but you have not yet named the steps, data, exceptions, or risks.

By the end, you should have a concrete Workflow Map that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Map — To map is to make the invisible steps of work visible in a clear order.
A map helps you see what happens before you choose a tool. It shows the inputs, outputs, handoffs, decisions, exceptions, and risks that are usually hidden inside a routine.
🏋 Try: Draw the five steps you follow before sending a weekly report.Mark the step where a mistake would be easiest to catch.

Work — Work means the actual tasks, decisions, conversations, files, approvals, and responsibilities that create a result.
Do not confuse work with a job title. Two people with the same title may do very different work, and AI affects tasks before it affects whole professions.
🏋 Try: List the twenty tasks you performed this week and group them by routine, judgment, relationship, and verification.Choose one task and write the result it is supposed to create.

Tool — A tool is any system, software, AI assistant, automation, template, or method used to help produce work.
A tool should serve the work problem. A tool is not automatically a solution; it becomes useful only when it improves quality, speed, access, safety, or learning without creating unacceptable risk.
🏋 Try: Name one tool you use daily and write what problem it actually solves.Identify one task where a checklist may be a better tool than AI.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

Name one real process you repeat often. Write the exact start and end point. Avoid a vague process like 'communication'; choose something observable such as 'turn client notes into a follow-up email.'
Break the process into small steps. Each step should take roughly 15 to 60 minutes and should produce a visible output.
Mark the data used at each step. Classify it as public, internal, confidential, personal, or highly sensitive.
Choose the technology role for each step: automate, augment, preserve, or redesign. Add one sentence explaining the choice.
Place human control at the point where a mistake can still be caught and corrected without serious harm.

Worksheet

Create a table or form with these fields and fill it as you work.

Process name and trigger
Step-by-step map
Data used at each step
Risk if the step is wrong
Best role for technology
Human checkpoint
Manual fallback

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Ask AI to automate the whole process now
Too early. You may automate hidden confusion, expose data, or miss the true risk point.

✅ Map the process first, then test one low-risk step
Best choice. You learn where the tool belongs before giving it too much control.

⚠️ Refuse all AI use because the process matters
Too broad. Important work can still include safe augmentation if control is designed well.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

I will describe a workflow. Help me break it into observable steps. For each step, ask what input, output, owner, data, exception, and risk I should record. Do not recommend tools yet.

Here is my workflow map: [paste]. Identify which steps are candidates for automation, augmentation, preservation, or redesign. Explain the risk behind each suggestion.

Act as a process quality reviewer. Challenge my workflow map. Where am I hiding judgment, sensitive data, or exceptions that a tool may mishandle?

Checkpoint

Can someone else understand your workflow map without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Create a one-page map of a real workflow and select one step for a two-week technology test. Save the baseline time, error type, and review method before testing.

Evidence to save

Your completed workflow map.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Choosing a tool before defining the work problem.
Treating a job title as one process instead of mapping actual tasks.
Putting human review only at the end, when the error is already hard to reverse.
Ignoring data sensitivity because the task feels routine.

🏁 Finish line. You are done when you have a usable Workflow Map, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.

Session 2 — Prompt With Context, Constraints, and Checks

⏱ 50 minutes · 🎯 You will build: three reusable prompts that produce useful drafts, summaries, and challenge questions.

Start here

You need help drafting, summarizing, or checking an idea. The first prompt you write is probably too vague, so the answer sounds confident but does not fit your context.

By the end, you should have a concrete Prompt Pack that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Prompt — A prompt is the instruction, context, material, and success criteria you give to an AI system.
A good prompt is structured delegation. It tells the tool what to do, for whom, with what limits, in what format, and how the answer should be checked.
🏋 Try: Rewrite 'summarize this' as a prompt that names audience, purpose, length, and key risks.Ask AI to list its assumptions after answering your prompt.

Context — Context is the situation surrounding the task: audience, purpose, background, constraints, available information, and decision use.
Without context, AI fills gaps with generic patterns. Context helps the output fit the real reader, organization, risk level, and action needed.
🏋 Try: Before asking for a draft, write who will read it and what decision they must make.Add two constraints from your real setting, such as low bandwidth, legal review, or limited time.

Constraints — Constraints are limits the answer must respect, such as length, tone, data rules, budget, audience, source boundaries, or format.
Constraints turn a broad request into useful work. They also stop the tool from inventing scope, tone, or authority it should not have.
🏋 Try: Tell AI to use only the facts you provide and to flag missing information instead of inventing it.Ask for a 150-word version, a bullet version, and a version for a beginner.

Checks — Checks are the verification steps used to inspect an AI output before trusting or using it.
Checks convert AI from a confident generator into a supervised assistant. They may include source review, recalculation, expert review, field validation, or stakeholder feedback.
🏋 Try: After an AI summary, compare three claims against the original document.Ask a colleague to review the output for missing context before sending it.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

Choose one real task. Write the weak prompt you would normally use in a hurry.
Add the missing context: audience, purpose, source material, constraints, tone, length, and decision use.
Tell the tool what not to do. Name forbidden sources, confidential details, claims it must avoid, and assumptions it should flag.
Ask for a checkable format. Request headings, bullets, a table, or a decision note only if that format helps you inspect the answer.
Add a verification instruction: sources to open, numbers to recalculate, risks to flag, or expert review needed.

Worksheet

Create a table or form with these fields and fill it as you work.

Task
Audience
Context
Source material
Constraints
Output format
Verification method
What the tool must not do

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Make the prompt longer until the result improves
Length alone is not quality. Add the right information, not every possible detail.

✅ Use a prompt frame and define checks before trusting the output
Best choice. Good prompting is structured delegation plus supervision.

⚠️ Copy a famous prompt from the internet
Sometimes useful, but it may not match your data, risk, audience, or work standards.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

Use this frame to help me write a better prompt: role, task, audience, context, constraints, output format, risks, and verification. Ask me one question at a time until the prompt is clear.

Rewrite this prompt so the output is easier to verify: [paste weak prompt]. Keep the task realistic and include a section for assumptions and limits.

After answering my prompt, create a verification checklist for the answer. Separate facts to check, assumptions to test, and judgment calls I must make.

Checkpoint

Can someone else understand your prompt pack without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Build a personal prompt pack with one drafting prompt, one summarizing prompt, and one critical review prompt. Test each prompt on a low-risk task and revise it once.

Evidence to save

Your completed prompt pack.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Asking for a final answer when you need a draft.
Forgetting to state the audience and decision context.
Letting the tool invent sources or policies.
Not asking for limits, assumptions, and verification steps.

🏁 Finish line. You are done when you have a usable Prompt Pack, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.

Session 3 — Delegate Drafting Without Losing Your Voice

⏱ 60 minutes · 🎯 You will build: a polished communication that started as an AI draft but ends in your own voice.

Start here

You need to write a message, update, article, report section, or announcement. AI can help you start, but the first draft sounds generic, overconfident, or unlike you.

By the end, you should have a concrete Draft Revision Log that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Delegate — To delegate is to give a task to another person or tool while keeping responsibility for the result.
Delegation to AI is not abandonment. You decide the task, define boundaries, review the output, correct errors, and remain accountable.
🏋 Try: Delegate only the first draft of a message, then revise facts and tone yourself.Ask AI to create three structure options, not the final decision.

Drafting — Drafting is creating a first version that will be reviewed, corrected, and improved.
A draft is raw material. AI is often useful at this stage because a rough structure can reduce blank-page friction, but the draft still needs human judgment.
🏋 Try: Ask AI for a first draft, then mark every sentence that needs verification.Create a before-after revision log showing what you changed.

Voice — Voice is the recognizable style, tone, values, and level of care in your communication.
Voice protects trust. A generic AI tone may sound smooth but fail to match the relationship, culture, emotion, or responsibility of the moment.
🏋 Try: Replace three generic phrases in an AI draft with words you would actually say.Add one real example from your context to make the message yours.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

Select a low-risk communication. Write the purpose in one sentence and name the reader's real concern.
Ask AI for a first draft only after you provide audience, tone, facts, length, and what must be avoided.
Highlight every claim that needs verification: names, numbers, dates, promises, policy statements, and cause-effect claims.
Rewrite the draft in your voice. Add one specific example, one honest limit, and one clear next step.
Compare the raw AI draft with your final version. Write down what you changed and why.

Worksheet

Create a table or form with these fields and fill it as you work.

Reader
Reader concern
Message purpose
Facts that must be accurate
Tone words
Sentences I changed
My added example
Final next step

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Send the polished AI draft because it sounds professional
Risky. Polished language can hide errors, exaggeration, or a voice that weakens trust.

✅ Use the draft as raw material and revise the facts, tone, and examples
Best choice. You gain speed while keeping responsibility.

⚠️ Avoid AI because writing must be fully human
Not necessary. You can delegate first drafts while preserving judgment and voice.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

Draft a [message type] for [audience]. Purpose: [purpose]. Use these facts only: [facts]. Tone: [tone]. Avoid exaggeration, invented details, and promises I did not make.

Review this draft for voice, clarity, and trust. Mark sentences that sound generic, inflated, unclear, or risky. Do not rewrite yet.

Help me create a revision log. Compare my raw draft and final draft, then list what changed in facts, tone, structure, specificity, and responsibility.

Checkpoint

Can someone else understand your draft revision log without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Produce one real communication using a draft-revise-verify cycle. Keep the raw AI output, your final version, and a short note explaining your human edits.

Evidence to save

Your completed draft revision log.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Letting AI decide the message's moral weight.
Keeping generic phrases because they sound smooth.
Failing to add lived context, examples, or constraints.
Editing only grammar while leaving weak thinking intact.

🏁 Finish line. You are done when you have a usable Draft Revision Log, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.

Session 4 — Verify AI Outputs and Sources

⏱ 60 minutes · 🎯 You will build: a verification plan for one AI-supported answer or recommendation.

Start here

An AI answer looks useful and confident. You need to decide whether it is good enough for a draft, a recommendation, or a decision that affects people, money, rights, safety, or reputation.

By the end, you should have a concrete Verification Plan that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Verify — To verify is to check whether something is accurate, supported, current, relevant, and safe enough to use.
Verification is proportional to risk. A low-risk brainstorm may need a light check; a decision affecting money, rights, safety, or reputation needs stronger evidence.
🏋 Try: Open the original source behind one AI citation and confirm it supports the claim.Recalculate a number from an AI output using a spreadsheet or calculator.

AI Outputs — AI outputs are the text, summaries, recommendations, classifications, images, code, tables, or actions produced by an AI system.
Outputs may be useful without being reliable. They can contain errors, bias, missing context, outdated information, or invented details.
🏋 Try: Break one AI answer into facts, assumptions, recommendations, and uncertainties.Highlight which parts of the output you can verify and which parts require expert judgment.

Sources — Sources are the original documents, data, people, observations, or records that support a claim.
A source is not just a link. It must be relevant, credible, current, and actually connected to the claim being made.
🏋 Try: Compare an AI summary with the original policy text.Check author, date, scope, and quoted passage before using a source.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

Classify the use as low, medium, or high impact. The higher the impact, the stronger the verification must be.
Separate the output into claims, calculations, recommendations, and assumptions.
Choose the verification lane for each part: primary source, calculation tool, expert review, field check, or user testing.
Open original sources instead of trusting citations or summaries. Check author, date, scope, and whether the source actually supports the claim.
Write a decision note that says what you verified, what remains uncertain, and who accepts responsibility.

Worksheet

Create a table or form with these fields and fill it as you work.

AI output
Impact level
Claims to verify
Numbers to recalculate
Assumptions to test
Missing context
Responsible reviewer
Decision after verification

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Trust the answer because it includes citations
Not enough. Citations can be wrong, irrelevant, invented, or misread.

✅ Verify only the parts that affect the decision
Best choice. Match verification effort to risk and impact.

⚠️ Reject all AI answers unless a human expert wrote them
Too rigid. AI can support work if verification is proportional and explicit.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

Break this answer into factual claims, calculations, recommendations, and assumptions. For each item, suggest the most appropriate verification method.

Act as a skeptical reviewer. What would make this answer false, misleading, incomplete, outdated, or unsafe?

Create a verification log for this output. Columns: item checked, method, evidence, result, remaining uncertainty, reviewer.

Checkpoint

Can someone else understand your verification plan without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Take one AI answer you planned to use. Verify it with at least two methods and write a short note on what changed after verification.

Evidence to save

Your completed verification plan.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Confusing fluent writing with evidence.
Checking only the final answer instead of the underlying assumptions.
Using the same AI tool to verify its own unsupported claim.
Skipping verification because the answer matches what you hoped was true.

🏁 Finish line. You are done when you have a usable Verification Plan, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.

Session 5 — Build a Safe Learning Sandbox

⏱ 45 minutes · 🎯 You will build: a safe sandbox charter for experimenting with AI or automation.

Start here

You want to learn by trying tools, but real customer data, staff information, financial files, or operational systems would make careless experimentation dangerous.

By the end, you should have a concrete Sandbox Charter that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Safe — Safe means the activity limits harm to people, data, operations, reputation, and legal or ethical obligations.
Safety does not mean zero risk. It means risks are known, bounded, monitored, and acceptable for the learning purpose.
🏋 Try: Use fictional customer profiles instead of real customer data.Set a stop rule before starting an experiment.

Learning — Learning is the process of turning experience into better skill, judgment, and future action.
Learning is stronger when it produces evidence. A sandbox should answer a specific question, not simply create impressions.
🏋 Try: Write one question your experiment must answer.After the test, record what worked, what failed, and what you would change.

Sandbox — A sandbox is a controlled space for testing ideas, tools, and workflows without exposing real systems to unnecessary risk.
The sandbox gives freedom inside boundaries. It protects sensitive data and operations while allowing hands-on practice.
🏋 Try: Test a tool with ten fictional cases before using real approved cases.Limit access, budget, and time for a two-week experiment.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

Write one learning question. Make it narrow, such as 'Can AI summarize public meeting notes accurately enough for a draft?'
Select safe data: fictional, public, anonymized, or explicitly approved.
Limit the sandbox. Define time, budget, users, permissions, and tools.
Choose success criteria and stop criteria before the test starts.
Capture lessons in a simple log so learning becomes transferable, not just personal curiosity.

Worksheet

Create a table or form with these fields and fill it as you work.

Learning question
Allowed data
Forbidden data
Approved tools
Permissions
Success criteria
Stop criteria
Lesson log

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Experiment freely with whatever files are easiest to access
Unsafe. Convenience is not a data governance rule.

✅ Use bounded tests with safe data and clear stop points
Best choice. You create space to learn without creating unnecessary exposure.

⚠️ Wait until every policy is perfect before learning anything
Too slow. A careful sandbox can produce learning while formal rules evolve.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

Help me design a two-week AI sandbox for this learning question: [question]. Include allowed data, forbidden data, permissions, success criteria, stop criteria, and lesson capture.

Review my sandbox charter for privacy, security, quality, and scope risks. Suggest tighter boundaries without killing useful learning.

Create five fictional test cases that resemble my real work without using confidential or personal data. My context is: [context].

Checkpoint

Can someone else understand your sandbox charter without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Run a two-week sandbox on fictional or public data. At the end, write a one-page lesson note: what worked, what failed, what risk appeared, and what you would test next.

Evidence to save

Your completed sandbox charter.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Using real sensitive data because the test is informal.
Testing many questions at once and learning nothing clearly.
Forgetting to define when the experiment should stop.
Keeping lessons private instead of turning them into shared practice.

🏁 Finish line. You are done when you have a usable Sandbox Charter, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.

Session 6 — Automate One Simple Workflow

⏱ 70 minutes · 🎯 You will build: a small automation pilot plan with baseline, test cases, and manual fallback.

Start here

A repetitive task drains time. You want to automate it, but the task may contain hidden exceptions, bad inputs, or review effort that cancels the benefit.

By the end, you should have a concrete Automation Pilot Card that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Automate — To automate is to let a system perform a task or part of a task with reduced manual effort.
Automation is best for stable, repeated, rule-based work. It becomes risky when the task contains hidden judgment, unclear inputs, or high-impact exceptions.
🏋 Try: Automate file naming for standard documents but keep human review for unusual cases.Measure the time spent correcting automation errors before calling it a success.

Simple — Simple means the task is narrow, understood, repeated, measurable, and low enough risk to test safely.
Simple does not mean unimportant. It means the first pilot is small enough to learn from without creating major disruption.
🏋 Try: Choose a weekly status summary before automating a complex approval process.Test ten cases instead of changing the workflow for everyone.

Workflow — A workflow is a sequence of steps, roles, inputs, outputs, decisions, and handoffs that produces a result.
Automation changes workflows, not isolated buttons. To improve a workflow, you must understand what happens before, during, and after the automated step.
🏋 Try: Write the current workflow before changing it.Name the manual fallback if the automated step fails.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

List five repetitive tasks and score each one for frequency, stability, risk, data readiness, and review effort.
Pick the safest high-value candidate, not the most exciting one.
Measure the current baseline: time, errors, rework, stress, and handoffs.
Design a small test with real but approved examples and a manual fallback.
Measure net gain after setup, correction, approval, and communication.

Worksheet

Create a table or form with these fields and fill it as you work.

Candidate tasks
Frequency score
Stability score
Risk score
Data readiness
Baseline time
Fallback method
Net result

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Automate the most annoying task immediately
Maybe, but annoyance does not prove stability or safety.

✅ Automate one stable, low-risk task and measure the full cycle
Best choice. You learn from a contained pilot.

⚠️ Automate only after the entire process is redesigned
Sometimes needed, but it can delay safe learning on a simple subtask.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

Here are five tasks: [list]. Score them for automation suitability using frequency, stability, risk, data readiness, and review effort. Explain the safest first pilot.

Help me create a baseline measurement plan for this task: [task]. Include time, errors, rework, handoffs, and user experience.

Design a manual fallback for this automation pilot. What should happen if the tool fails, produces uncertainty, or creates a suspicious result?

Checkpoint

Can someone else understand your automation pilot card without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Run a small automation pilot on one task. Keep a before-after log for at least ten cases and decide whether to continue, change, or stop.

Evidence to save

Your completed automation pilot card.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Automating an unstable process.
Ignoring setup, review, and correction time.
Failing to test exceptions.
Removing the manual fallback too early.

🏁 Finish line. You are done when you have a usable Automation Pilot Card, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.

Session 7 — Use Data Without Fooling Yourself

⏱ 65 minutes · 🎯 You will build: a data-question checklist for judging one metric before using it in a decision.

Start here

A dashboard, report, or AI analysis gives you a number. The number may be useful, but it may also hide missing data, biased samples, bad definitions, or a misleading chart.

By the end, you should have a concrete Metric Interrogation Sheet that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Data — Data is recorded information, such as numbers, text, categories, dates, images, audio, transactions, or observations.
Data is shaped by how it is collected, labeled, cleaned, excluded, and interpreted. It is never neutral simply because it is numerical.
🏋 Try: Ask who is missing from a dataset before using it.Define a metric in plain language and check if others define it the same way.

Fooling Yourself — Fooling yourself means drawing a confident conclusion from weak, incomplete, biased, or misunderstood evidence.
AI and dashboards can make weak reasoning look precise. You avoid self-deception by checking definitions, samples, missing context, correlation, and incentives.
🏋 Try: Write three alternative explanations for a trend before deciding what caused it.Add one qualitative signal to explain a dashboard number.

Metric — A metric is a measurement used to describe performance, behavior, quality, cost, risk, or progress.
A metric should improve a decision. If nobody knows what decision the metric supports, it may create noise rather than clarity.
🏋 Try: Connect one metric to one decision it should improve.Add a second metric that could reveal harm hidden by the first.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

Choose one metric you often see or use. Write the decision it influences.
Define the metric in plain language. If two people define it differently, stop and resolve that first.
Ask who or what is missing from the data. Look for silent exclusions.
Check whether the chart implies a cause when it only shows a pattern.
Add a second signal that would confirm, challenge, or explain the metric.

Worksheet

Create a table or form with these fields and fill it as you work.

Metric
Decision influenced
Definition
Data source
Who is missing
Possible bias
Second signal
Decision rule

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Use the metric because it is already on the dashboard
Weak choice. Availability is not validity.

✅ Question the metric before connecting it to action
Best choice. You protect decisions from false precision.

⚠️ Reject metrics because they can be biased
Too far. Metrics are useful when interpreted with context and humility.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

Interrogate this metric: [metric]. Ask questions about definition, source, sample, missing data, bias, correlation, causality, and decision relevance.

This chart seems to show [interpretation]. Give me three alternative explanations and one additional data point that would help decide between them.

Help me rewrite this dashboard insight so it separates observation, interpretation, uncertainty, and recommended action.

Checkpoint

Can someone else understand your metric interrogation sheet without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Take one metric from your work or studies and create a one-page metric note. Include definition, risk of misreading, second signal, and how you will use it responsibly.

Evidence to save

Your completed metric interrogation sheet.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Treating a precise number as a true number.
Ignoring who is missing from the sample.
Confusing correlation with causation.
Using a metric without naming the decision it should improve.

🏁 Finish line. You are done when you have a usable Metric Interrogation Sheet, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.

Session 8 — Work With Multimodal AI

⏱ 60 minutes · 🎯 You will build: a two-mode AI test comparing text, image, audio, video, or document inputs.

Start here

The same problem can arrive as text, photo, voice note, PDF, spreadsheet, video, or form. Each format changes accuracy, privacy, accessibility, and review effort.

By the end, you should have a concrete Multimodal Test Card that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Multimodal — Multimodal means using more than one type of input or output, such as text, image, audio, video, tables, or documents.
Different modes reveal different information and create different risks. A photo, voice note, and spreadsheet do not need the same checks.
🏋 Try: Compare an AI summary from a voice note with one from written notes.Check whether an image-based answer missed details visible to a human.

AI — AI refers to systems that perform tasks associated with prediction, generation, classification, pattern recognition, or decision support.
AI can assist with perception and language, but it does not automatically understand responsibility, context, or consequences.
🏋 Try: Use AI to organize observations, then decide yourself what action is responsible.Ask the tool to state uncertainty and what original material a human should inspect.

Mode — A mode is the form in which information enters or leaves a system, such as speech, image, text, or structured data.
Choosing the right mode affects accessibility, accuracy, privacy, and review effort.
🏋 Try: Test whether a text form or voice note gives clearer information for field reports.Use a checklist to decide whether a photo contains personal or sensitive information.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

Choose one problem that appears in more than one format, such as support requests, maintenance notes, interview feedback, or training questions.
Select two input modes to compare. Use approved or fictional data.
Define what a good output must include and what would make it unsafe.
Run the same task in both modes and compare accuracy, missing context, privacy, and review time.
Decide which mode is useful, which needs control, and which should not be used for this task.

Worksheet

Create a table or form with these fields and fill it as you work.

Problem
Mode 1
Mode 2
Expected output
Accuracy result
Privacy concern
Review effort
Best use

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Use the mode that feels most impressive
Not enough. Impressive input handling does not prove reliable output.

✅ Compare modes using the same task and clear criteria
Best choice. You learn which mode actually supports the work.

⚠️ Avoid non-text AI because it is too risky
Too broad. Some multimodal uses are safe when data and review are controlled.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

Help me design a comparison test for this problem: [problem]. I want to compare [mode 1] and [mode 2]. Include criteria for accuracy, privacy, accessibility, and review effort.

Create a review checklist for AI output based on [photos/audio/documents/video]. Include what original material a human must inspect before acting.

Generate fictional sample inputs for a multimodal AI test in this context: [context]. Avoid personal or confidential data.

Checkpoint

Can someone else understand your multimodal test card without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Run a two-mode test and write a short recommendation: use, use with controls, or do not use. Include evidence from at least five sample cases.

Evidence to save

Your completed multimodal test card.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Forgetting that images, voices, and documents can contain sensitive data.
Judging output quality without checking the original input.
Using the same review criteria for every mode.
Letting accessibility benefits hide privacy or accuracy problems.

🏁 Finish line. You are done when you have a usable Multimodal Test Card, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.

Session 9 — Protect Data, Privacy, and Cybersecurity

⏱ 75 minutes · 🎯 You will build: a data boundary card for one AI use case.

Start here

You want speed, but the information you handle may include confidential, personal, regulated, contractual, or security-sensitive material.

By the end, you should have a concrete Data Boundary Card that you can use in your work, studies, team, or personal development. Do not only read this page. Open a blank note, document, or worksheet and complete each action before moving on.

Title concepts to master

Before you start the actions, make sure the main words in the title are practical, not abstract. Use the definitions, explanations, and examples below as a mini-warm-up.

Data Protection — Data protection means controlling how information is collected, used, shared, stored, retained, and deleted.
Protection starts before the prompt. You must know what kind of data you have and which systems are approved to process it.
🏋 Try: Classify data before pasting any example into an AI tool.Remove names, account numbers, and identifying details from test cases.

Privacy — Privacy is the right and expectation that personal information is handled with care, purpose, limits, and respect.
Privacy is not only secrecy. It also includes consent, relevance, minimization, access control, and fair use.
🏋 Try: Ask whether a person's information is necessary for the AI task.Replace real profiles with fictional profiles for experimentation.

Cybersecurity — Cybersecurity is the practice of protecting systems, accounts, data, and operations from unauthorized access, misuse, disruption, or attack.
AI can expand the attack surface through connected tools, files, prompts, plugins, accounts, and hidden instructions.
🏋 Try: Limit an AI assistant's access to only the files needed for the task.Treat external documents as untrusted inputs until checked.

Quick self-check

Where does this topic already appear in your work or life?
What mistake would be costly if you handled this topic casually?
What proof would show that you improved by the end of this session?

Do this now

List the data types involved in one AI use case.
Classify each data type: public, internal, confidential, personal, or highly sensitive.
Name which tools are approved for each class. If you do not know, mark it as unknown instead of guessing.
Apply least privilege: decide the minimum access, retention, and sharing needed.
Write the incident path: what to do if data is pasted into the wrong tool, exposed, or used in an unsafe output.

Worksheet

Create a table or form with these fields and fill it as you work.

Use case
Data type
Data class
Approved tool
Allowed action
Forbidden action
Access limit
Incident contact

Choose your path

Read the options. Pick the one you would naturally choose, then check the consequence.

⚠️ Paste the data because the task is urgent
Dangerous. Urgency does not erase privacy, security, or contractual duties.

✅ Classify the data first and use only approved paths
Best choice. It protects trust while still allowing useful work.

⚠️ Never use AI for any internal work
Too broad. Many internal uses are safe when data boundaries are clear.

Prompts you can use

Use these prompts only with information you are allowed to share. Replace the bracketed parts with your own context.

Help me classify the data in this AI use case: [describe]. Ask me about public, internal, confidential, personal, and highly sensitive information. Do not ask me to paste real sensitive data.

Create a data boundary card for this use case. Include allowed tools, forbidden data, access limits, retention concerns, and incident steps.

Act as a cybersecurity reviewer. What could go wrong if an AI assistant is connected to these files or tools: [describe connection]?

Checkpoint

Can someone else understand your data boundary card without extra explanation?
Did you separate facts, assumptions, preferences, and decisions where relevant?
Did you name the human responsibility, not only the tool or technique?
Did you protect confidential, personal, or sensitive information?
Is the next action small enough to do within seven days?

Small project

Create three data boundary cards for tasks you or your team perform. Share them with the person responsible for policy, security, or operations.

Evidence to save

Your completed data boundary card.
One before-and-after note showing what changed because of the tutorial.
One risk, limit, or open question you discovered.
One next action with a date.

Common mistakes to avoid

Assuming a tool is safe because it is popular.
Using real personal or confidential data in informal tests.
Giving a connected assistant more access than the task requires.
Not knowing who to contact when something goes wrong.

🏁 Finish line. You are done when you have a usable Data Boundary Card, one decision about what to do next, and one piece of evidence that shows your thinking became clearer or safer.