EKALAVYA ACADEMY | SHORT BOOKS

When not to use AI, and who stays in charge

How to decide which jobs stay with people, make human oversight real and keep accountability simple enough to last.

Ekalavya Academy by Almost Magic
Short Book 10 of 10 | Free to read and share

How to read this book

Before You Start

1. Learn
A few slides explain one idea in full sentences.

2. See it
A short, made-up example gives the idea a face.

3. Try it
A ten-minute exercise in the tools you already use.

4. Check it
Three questions. Guidance is at the back, not next to the question.

5. Keep it
A four-sentence summary to come back to.

Two levels in every chapter. The main slides are written so a first-time reader can follow them. A "Going deeper" slide gives the same idea at the level an experienced reader would argue about.

This book works with any AI tool you already have. It never asks you to buy or sign up for a particular product.

Where this book sits in the series

The Plan

Book 1	What AI is, and what it is not	Foundations
Book 2	Productivity with AI: writing, summaries, research, meetings	Practical use
Book 3	From one task to a repeatable workflow, and AI deputies	Workflow automation, agents
Book 4	Handling data safely	Data handling
Book 5	When a data shortcut becomes the rule	Data quality and governance
Book 6	Staying secure with AI	Security
Book 7	Rules, risk and what you must know	Risk and compliance awareness
Book 8	Your AI rules on one page	Policy and accountability
Book 9	When AI gets it wrong	Recovering from AI mistakes
Book 10	When not to use AI, and who stays in charge	Judgement and oversight

Each book works on its own. Read them in order or pick the one you need.

In this book

Contents

Chapter 1
Red, amber, green

Chapter 2
Human oversight that works

Chapter 3
Accountability in plain words

Chapter 4
Saying no, and the hard cases

At the back
Guidance and answers for every question and exercise

Red, amber, green

Chapter 1

By the end of this chapter you will be able to:

Judge the job rather than the tool.

Sort uses into green, amber and red.

Name the jobs that stay with people.

Know when checking costs more than doing.

Time: about 40 minutes, including the exercise.

Judge the job, not the tool

Learn

AI is useful for many jobs and wrong for others. The same tool can sit on either side.

Book 7 asks who is affected, how serious a mistake is and whether it can be undone. This chapter turns those questions into three simple colours.

Saying "not for this job" is part of using AI well, not a failure of it.

IN PLAIN WORDS

Knowing when not to use it is a skill.

Three colours

Learn

Green
Low stakes, easy to check, no sensitive data. Use freely.

Amber
A customer will see it, figures are involved, or a mistake costs time or money. Use with a named person's check.

Red
Rights, money, safety or sensitive data are at stake, or you cannot check the result. A person decides.

Review
Colours move. Look again when the data, the audience or the tool changes.

Questions before you start

Learn

Can I check it?
If you cannot tell right from wrong, you cannot rely on the result.

What if it is wrong?
Annoying, costly, or harmful to a person?

Who is affected?
Customers, staff, applicants or the public.

Can it be undone?
A draft can. A sent message or a decision about a person often cannot.

Jobs that stay with people

Learn

Some jobs belong to a person because someone must answer for the result.

Decisions about hiring, firing, pay, credit, discipline or health. Delivering bad news or handling a distressed customer. Anything you could not explain to the person it affects. Final sign-off on legal, safety or financial matters.

AI can prepare material for these. A person decides.

RULE OF THUMB

If you cannot explain it to the person, do not hand it over.

When checking costs more than doing

Going Deeper

The checking test
If checking takes as long as doing the job, AI has saved nothing.

Long outputs
A long result is harder to check well than a short one. Cut the scope.

Skills fade
Hand over a skill for good and you can no longer judge the result.

Pressure to use it
A tool being available is not a reason to use it on this job.

A short example

See It

Lena runs a six-person dental clinic. She uses AI to draft appointment reminder texts (green) and to summarise supplier emails with a staff check (amber). For now she uses it for nothing in patient notes or complaints (red).

A new receptionist asked why. Lena showed her the colour list on the staff board.

"We use it where a wrong answer is cheap to catch," she said.

THE LESSON

A short colour list answers most "can I?" questions.

This is a made-up example for teaching.

Sort your uses

Try It

1. List five ways you use, or want to use, AI.

2. Ask the four questions of each.

3. Mark each one green, amber or red.

4. For every red, write who decides instead.

YOUR TOOLS

Use the quick risk screen from Book 7 if you want a second view.

GUIDANCE

What to look for is in the Guidance section at the end.

Check yourself

Check It

Question 1
Why judge the job rather than the tool?

Question 2
Give one example each of a green, amber and red use.

Question 3
When has AI saved you nothing, even if the output is good?

Answer in your own words before you read the Guidance section at the end.

Chapter 1 in four sentences

Keep It

KEEP THIS

The same tool can be right for one job and wrong for another. Sort uses into green, amber and red by what is at stake and whether you can check the result. Decisions about people, and anything you cannot explain, stay with a person. If checking costs as much as doing, skip the tool.

Human oversight that works

Chapter 2

By the end of this chapter you will be able to:

Say what real oversight is.

Choose the right level of human involvement.

Make sure a person can pause or override.

Spot rubber-stamping.

Time: about 40 minutes, including the exercise.

Oversight is not a rubber stamp

Learn

"A person looks at it" sounds safe. It is only safe if the person has time, knowledge and the power to say no.

Book 3 made the same point about approval: clicking through a long list without reading is not approval.

National AI Centre guidance asks for human control in proportion to the AI's autonomy and the stakes. Small stakes, light oversight. Big stakes, a person in charge.

REMEMBER

Oversight means someone can say no.

Four roles for the human

Learn

Approves each one
Checks every output before it takes effect. For high stakes.

Samples and reviews
Checks a share of results and the log. For steady, lower-risk work.

Sets limits and can stop it
Defines what it may do and can pause it. The minimum for anything automated.

Not involved
Only for trivial, reversible jobs. Never for decisions about people.

Match oversight to stakes

Learn

Low stakes
Spot checks, and a way for people to report a problem.

Medium stakes
A named person checks before it reaches a customer.

High stakes
A person decides. AI prepares and a person approves each case.

Any automation
A pause control, a log and a named person who reads it (Book 3).

Make the override real

Learn

A pause button nobody can reach is not a control. Check four things.

Who can stop it, and do they know they can? Can they do it quickly, without asking first? Will they hear about problems in time? Do they have the skill to judge the result?

Test it once. Ask a colleague to pause the flow while you watch.

RULE OF THUMB

Test the stop button before you need it.

Why oversight fails

Going Deeper

Automation bias
People trust a confident machine and stop questioning it.

Reviewer fatigue
A long queue of "fine" outputs trains the eye to skim. Sample and rotate.

No time
If the check is not in the schedule, it will not happen.

No authority
A junior person who cannot say no is not oversight.

A short example

See It

Arjun runs a small accounting firm with an AI flow that drafts reminder emails to clients. A staff member was approving forty drafts a day with a single click.

In a review, Arjun found two with the wrong client name. He changed the setup: the staff member now reads five random drafts in full, the flow pauses itself on any amount over a limit, and Arjun reads the weekly log.

The daily check took less time, and it caught more.

THE LESSON

Real oversight is designed, not assumed.

This is a made-up example for teaching.

Test an override

Try It

1. Pick one AI use that runs with little supervision.

2. Write who can pause it, and how.

3. Ask them to try it, and time how long it takes.

4. Fix any gap you find: access, skill or time.

YOUR TOOLS

A ten-minute test with a colleague is enough.

GUIDANCE

What to look for is in the Guidance section at the end.

Check yourself

Check It

Question 1
What makes human oversight real rather than a rubber stamp?

Question 2
What level of oversight suits a decision about a person?

Question 3
Why test the pause control before you need it?

Answer in your own words before you read the Guidance section at the end.

Chapter 2 in four sentences

Keep It

KEEP THIS

A person looking is not oversight unless they have time, knowledge and the power to say no. Match oversight to the stakes: approve each one, sample, or set limits and keep a stop. Make the override reachable and test it. Watch for skimming and over-trust.

Accountability in plain words

Chapter 3

By the end of this chapter you will be able to:

Explain governance in plain words.

Name an owner for AI use and for each use.

See how the series fits together.

Keep governance small enough to last.

Time: about 40 minutes, including the exercise.

Governance in plain words

Learn

Governance sounds grand. For a small business it is four answers: who decides, what is allowed, how it is checked, and what happens when it goes wrong.

Book 5 used the word for data. The same idea applies to AI use. One person owns it. Each use has a named person. There are a few written rules and a regular look at how it is going.

The National AI Centre guidance in Book 7 describes the same pattern, scaled to your size.

IN PLAIN WORDS

Governance is knowing who is responsible.

How the books fit together

Learn

Who is responsible
Books 1 and 8: a named person owns each result and each decision.

What is allowed
Book 8, the one-page rule sheet. Book 7, the rules that may apply.

How data is handled
Books 4 and 5: labels, owners, a record for each data set.

How it stays secure
Book 6: accounts, tools, scams and attacks on AI tools.

How risk is judged
Book 7 and this book: the risk screen, red, amber, green.

When it goes wrong
Book 4, Chapter 4, and Book 9: first hour, correction, learning.

Four questions for every AI use

Learn

Who owns it?
One named person, with cover when they are away.

What data goes in?
Use the labels from Book 4.

How is it checked?
By whom, against what, and how often?

How is it stopped?
Who can pause it, and how fast?

Small business, small governance

Learn

You do not need a committee. You need a page and a habit.

Start with the owner, the rule sheet and a register of AI uses (Book 7). Add a review each quarter. Add more only when your use grows or the risk rises.

The guidance itself says to start small and build up. Keep a note of what you did, so you can show it.

RULE OF THUMB

A page and a habit beat a binder.

Governance as a habit

Going Deeper

Owner on leave
Name who covers, or decisions stall or happen by default.

Drift
Rules written once go stale. Put the review in the calendar.

Evidence
Dates, names and decisions show you did what you said.

Proportion
More rules for riskier uses, a light touch for green ones.

A short example

See It

Ngaire runs a ten-person landscaping firm. She was told she "needed AI governance" and pictured a binder.

She wrote one page: who owns AI use (her), the four questions for each use, and a list of six uses with owners. She put a review in the calendar for the first Monday of each quarter.

At the first review she dropped two uses and added one rule.

THE LESSON

Governance is a page, a name and a calendar entry.

This is a made-up example for teaching.

Write your governance page

Try It

1. Name the owner of AI use in your business, and who covers.

2. List your AI uses, with an owner for each.

3. Answer the four questions for the riskiest one.

4. Put a quarterly review in your calendar.

YOUR TOOLS

One page, in a place the team can find.

GUIDANCE

What to look for is in the Guidance section at the end.

Check yourself

Check It

Question 1
What four answers make up governance in a small business?

Question 2
Why name who covers when the owner is away?

Question 3
What is the smallest set of things worth starting with?

Answer in your own words before you read the Guidance section at the end.

Chapter 3 in four sentences

Keep It

KEEP THIS

Governance is knowing who is responsible. Four answers cover it: who decides, what is allowed, how it is checked, and what happens when it goes wrong. Start with an owner, a rule sheet, a register and a review in the calendar. Keep it small enough to last.

Saying no, and the hard cases

Chapter 4

By the end of this chapter you will be able to:

Respond when a customer wants a person.

Settle common grey areas in advance.

Treat other people's work and confidential material with care.

Review your lines on a schedule.

Time: about 40 minutes, including the exercise.

When a customer wants a person

Learn

Someone will ask to speak to a human. They may not want their matter handled by a tool, or may have been let down by one.

Make it easy. Say how to reach a person, and do it without a fight. Tell people when AI is involved where it could mislead (Book 7).

An easy way out builds more trust than a clever tool.

RULE OF THUMB

Always keep a way to reach a person.

Grey areas to settle in advance

Learn

"It is faster"
Speed is not a reason to skip the check on an amber or red use.

"Everyone does it"
Other businesses carry risks you cannot see. Use your own colours.

"A client asked for it"
Check what the client allows, and tell them how you use AI.

"I used my own account"
Personal accounts put work data outside your control (Book 6).

Other people's work and material

Learn

Client material
Treat it as sensitive unless you have permission. See Book 4.

Other people's content
Pasting it into a tool, or publishing output close to it, can raise copyright questions.

Who owns the output
Rules differ by tool and country. Read the terms and ask an adviser when it matters.

Your sources
Say where facts came from. A tool cannot vouch for itself.

Say no with a reason

Learn

Sometimes the right answer to "can I use AI for this?" is no. Say it with a reason and an alternative.

Give the reason in one line: who it affects, or what it risks. Offer what they can do instead, such as a draft to edit by hand, or a person to ask.

A clear no with a reason is easier to follow than a vague rule.

IN PLAIN WORDS

A no needs a reason and an alternative.

Opting out is a feature

Going Deeper

Customers may decline
Offer an alternative where you can, and do not penalise them.

Staff may decline
Some people will not want to use it. Ask why before you push.

Review the line
What was red last year may be amber now. Look each quarter.

Write it down
Dates and reasons show your thinking if you are asked.

A short example

See It

Hana runs a small property management firm. A tenant emailed to say she did not want her maintenance request handled by a chatbot.

Hana replied with a phone number and a named person. She also added a line to the website: "Where we use AI, you can always ask for a person."

Two staff asked about using personal AI accounts for tenant emails. She said no, with a reason: tenant details are sensitive. She offered a business account instead.

THE LESSON

Easy exits and clear reasons keep trust.

This is a made-up example for teaching.

Write three answers

Try It

1. Write the line a customer sees about reaching a person.

2. Pick one grey area your team will meet, such as personal accounts.

3. Answer it in two sentences: yes or no, the reason, the alternative.

4. Add both to your rule sheet from Book 8.

YOUR TOOLS

Plain words. Read them aloud to check they sound human.

GUIDANCE

What to look for is in the Guidance section at the end.

Check yourself

Check It

Question 1
Why make it easy to reach a person?

Question 2
What goes in a good "no"?

Question 3
Why check the tool's terms before putting other people's work into it?

Answer in your own words before you read the Guidance section at the end.

Chapter 4 in four sentences

Keep It

KEEP THIS

Always keep a way to reach a person, and say so. Settle grey areas before they come up, and give a reason with every no. Treat client material and other people's work with care, and ask an adviser about ownership and copyright. Review your lines every quarter.

Guidance and answers

At The Back

Questions and exercises stay in the chapters, on their own slides. Guidance lives here, so you can try first and look afterwards.

Try the task or question in your own words, then compare. Where your answer differs, that is worth a note, not a correction.

WHERE TO FIND IT

The Guidance section below: answers and exercise guidance, one part for each chapter.

Chapter 1: answers and guidance

Guidance

Answer 1
Risk lives in the use, such as who is affected and how serious a mistake is, not in the tool itself.

Answer 2
Any sensible examples. For instance: green, tidy a draft. Amber, summarise a supplier email with a check. Red, decide who to hire.

Answer 3
When checking takes as long as doing the job, or you cannot judge the result.

Exercise
A good list has a named decider for every red use.

Chapter 2: answers and guidance

Guidance

Answer 1
The person has time, skill and authority, and can pause or override quickly.

Answer 2
A person decides, with AI preparing the material, and approves each case.

Answer 3
A control nobody can reach, or nobody dares use, fails when you most need it.

Exercise
A good test finds one gap and fixes it, such as access, skill or time.

Chapter 3: answers and guidance

Guidance

Answer 1
Who decides, what is allowed, how it is checked, and what happens when it goes wrong.

Answer 2
Without cover, decisions stall or happen by default.

Answer 3
An owner, a one-page rule sheet, a register of uses and a regular review.

Exercise
A good page names people, not only roles, and has a review date.

Chapter 4: answers and guidance

Guidance

Answer 1
It builds trust, and some people do not want a tool handling their matter.

Answer 2
A reason in one line and an alternative the person can act on.

Answer 3
Terms differ on what the tool keeps and how it may use what you enter. Other people's work may carry rights.

Exercise
A good answer says yes or no, why, and what to do instead.

About Ekalavya Academy

Find Out More

Ekalavya Academy is a learning series from Almost Magic Tech Lab. The books are written for people who teach themselves, using the tools they already have.

Sources used: National AI Centre, "Guidance for AI adoption: foundations" (ai.gov.au), as summarised in Book 7, on human control and starting small. Awareness, not legal advice.

Almost Magic builds tools that help people check and govern their AI use. The books never depend on those tools.

LINKS

Almost Magic: almostmagic.net.au

Case studies: ai-casestudies.almostmagic.net.au

Page for this book on the Academy site: link to be added when the page exists.

Ekalavya Academy by Almost Magic | almostmagic.net.au