Introduction — Nobody's Problem Is That They Cannot Write a Prompt
A session scored 4.7, somebody pasted a supplier contract into a public tool during the exercise, and three months later nothing had changed.
⚠ The trade sells prompting.
⚠ ⚠ Prompting is the least important thing in the room, and teaching it first produces confident misuse.
What Actually Fails
⚠ Walk into any organisation that has "started using AI" and the failures are in this order.
| ⚠ Rank | ⚠ Failure | |
|---|---|---|
| ⚠ 1 | ⚠ ⚠ Nobody knows what they are allowed to do | ⚠ ⚠ Near-universal |
| ⚠ 2 | ⚠ ⚠ Confidential material goes into unapproved tools | ⚠ ⚠ Very common |
| ⚠ 3 | ⚠ ⚠ Output is used without being checked | ⚠ ⚠ Very common |
| ⚠ 4 | ⚠ ⚠ People are quietly frightened and say nothing | ⚠ ⚠ Common, invisible |
| ⚠ 5 | ⚠ Nobody has changed a workflow | ⚠ ⚠ Common |
| ⚠ 6 | ⚠ Prompting is inefficient | ⚠ ⚠ Rarely the constraint |
⚠ ⚠ Notice what happens if you teach number six first
⚠ ⚠ ⚠ You produce forty enthusiastic people with better prompting skills, operating in an organisation that has not decided what is permitted, using tools nobody approved, on material nobody classified, checking nothing.
⚠ You have not improved anything. You have increased the throughput of an unmanaged process.
⚠ ⚠ ⚠ And the feedback will be excellent, because teaching prompting is genuinely satisfying and people enjoy it. Chapter 25.
The Order That Works
☐ ⚠ ⚠ What are we allowed to do? — Chapter 18
☐ ⚠ ⚠ What must never go in? — Chapter 7
☐ ⚠ ⚠ Why can we not trust the output? — Chapter 24
☐ ⚠ Which of our actual workflows is worth changing? — Chapter 21
☐ ⚠ How do we do that well? — Chapter 20
☐ ⚠ Who checks it, and when? — Chapter 24
⚠ ⚠ Prompting is step five
⚠ ⚠ ⚠ Not because it does not matter, but because it is the only step that is useless without the other five — ⚠ ⚠ ⚠ and dangerous without the first three.
The Four Rules
⚠ 1. You cannot vouch for the tool.
⚠ ⚠ You are recommending a system that produces confident false statements, changes without notice, and reflects biases nobody has fully characterised. ⚠ ⚠ ⚠ That is not a caveat at the end of the day. It is the first lesson. Chapter 4.
⚠ 2. You do not write their policy.
⚠ ⚠ ⚠ A policy is a governance document with employment, legal, contractual, and regulatory consequences. ⚠ ⚠ ⚠ You can facilitate the decisions and draft for their review. ⚠ ⚠ ⚠ Somebody accountable signs it, and it is never you. Chapter 5.
⚠ 3. The room is not the buyer.
⚠ ⚠ The person paying wants adoption. ⚠ ⚠ Some people in the room are wondering whether they are training their replacement. ⚠ ⚠ ⚠ You owe honesty to both, and you never report on individuals. Chapter 6.
⚠ 4. Everything they practise on is a real document.
⚠ ⚠ ⚠ Abstract exercises do not transfer, so good training uses real work — ⚠ ⚠ which makes your workshop a live data-exposure event unless you design it not to be. Chapter 7.
⚠ ⚠ How each breaks
| ⚠ Rule | ⚠ The sentence that breaks it |
|---|---|
| ⚠ 1 | ⚠ "Which tool should we buy?" |
| ⚠ 2 | ⚠ "Could you just draft us a policy while you're here?" |
| ⚠ 3 | ⚠ "Who in the session seemed resistant?" |
| ⚠ 4 | ⚠ "Just bring something you're working on." |
⚠ ⚠ ⚠ Every one is reasonable, friendly, and said by somebody paying you.
The Central Argument
⚠ You are not teaching a tool.
⚠ You are changing what forty people do with confidential information on Monday morning.
⚠ ⚠ Which means the deliverable is not a workshop
⚠ ⚠ ⚠ It is a small number of named workflows, changed deliberately, with a review step somebody owns, still in use ninety days later. Chapter 26.
⚠ ⚠ ⚠ Three workflows genuinely changed beats forty people who enjoyed a session.
⚠ ⚠ And the measurement problem that defines this trade
⚠ A workshop everybody enjoyed and nobody changed anything after is the standard outcome — and it is indistinguishable from success on the feedback form.
⚠ ⚠ ⚠ 4.7 out of 5 means people had a good afternoon. ⚠ ⚠ ⚠ It carries no information about whether anything is different. Section 25.1.
What This Book Refuses to Tell You
⚠ ⚠ This is a book about training people to use AI tools that will not name a single tool.
Deliberately.
⚠ ⚠ ⚠ Products change monthly. ⚠ ⚠ Capabilities change monthly. ⚠ ⚠ Data-handling terms differ by plan, by contract, by region, and by how the client's own administrator configured it.
⚠ A confident sentence about what a tool does with a client's data is the most dangerous thing you can say in this trade.
So this book states:
⚠ No product name, capability, limitation, or benchmark.
⚠ ⚠ No claim about what any tool does with input.
⚠ No law, regulation, data-protection obligation, employment rule, or copyright position.
⚠ No licensing requirement, insurance position, or tax rule.
⚠ No price, rate, or income figure.
⚠ ⚠ ⚠ Every one routes to the vendor's current documentation, the client's own contract, or a qualified professional — ⚠ ⚠ read by you, and dated. Resource 3.
Who This Is For
⚠ People who have used these tools seriously in real work and can explain things.
⚠ ⚠ Trainers and L&D people adding this to what they already do.
⚠ ⚠ ⚠ Consultants whose clients keep asking.
⚠ ⚠ ⚠ People inside organisations who have been asked to teach colleagues and want to do it properly.
⚠ Anybody who has run one of these and felt the gap between the feedback score and what actually happened.
⚠ Who it is not for
⚠ ⚠ Anybody who wants to sell prompt packs.
⚠ ⚠ ⚠ Anybody who wants to recommend a tool for a commission. Section 2.5.
⚠ ⚠ ⚠ Anybody who would draft a governance policy they are not qualified to write.
How the Book Is Built
| Part | Chapters | |
|---|---|---|
| ⚠ One | 1–7 | ⚠ The industry, the models, and the four rules |
| ⚠ Two | 8–16 | ⚠ Refusals, credibility, niche, plan, budget, compliance |
| ⚠ Three | 17–26 | ⚠ Assessment, policy gap, curriculum, prompting, examples, delivery, exercises, distrust, evaluation, follow-up |
| ⚠ Four | 27–32 | ⚠ Packages, prices, agreements, delivery, communication, quality |
| ⚠ Five | 33–37 | ⚠ First clients, being found, complaints, money, AI, and your first year |
A Word About AI
⚠ A hundred and forty-eight prompts, and a chapter about where AI must never go.
⚠ ⚠ In this trade the line is sharper than in most, because your credibility depends on it:
⚠ You are teaching people to check the output. If your own materials contain something you did not check, you have taught the opposite lesson at your own expense. Section 37.3.
⚠ ⚠ ⚠ Never put a client's documents, participant details, assessment findings, or internal information into an AI system. Section 37.2.
What Jules Learned
⚠ The finance contract came out of the tool's history when somebody finally looked.
⚠ ⚠ Nothing catastrophic followed, which is the usual outcome and is not the same as nothing having happened.
⚠ ⚠ ⚠ And the session had scored 4.7.
Two things followed, and they are this book:
⚠ Design the exercise so that a mistake in the room is impossible, because somebody will make one and it will be your design that decided what it cost.
⚠ And ask the question first that nobody in the room can answer — "what are we actually allowed to do?" — because until it has an answer, everything else you teach is throughput.
©2026 James Henderson / https://localhandyman.work