What are you wondering about?

Start by asking yourself what you would do if AI were not available. What would you read? Which questions would you need answered? What information would you dig out from earlier stages of the project?

Gather the material you would need to do the work well. Put your initial thinking into a sketch: what you want to achieve, what you already know, how you would approach it and where you need to find out more. Bring your experience into that preparation. It gives the collaboration a direction that comes from you.

There are no magic prompts. What we practise here are thinking patterns. Once you have a sketch, you can invite AI to ask questions that help it understand the project. For example, upload your sketch and say:

Help us get oriented

This is my initial thinking about a project we are going to work on. Give me 10 questions about the project so that my answers will help you get better oriented and help keep the project uniquely mine.

Use your answers to develop the shared understanding before moving on to proposals. You can adapt the wording; the preparation and the choices are what matter.

Link to this answer

The highest priority is to make conscious decisions based on solid reasoning. If AI suggests a promising alternative, ask it to compare that idea with your initial plan against the aims and constraints of the project.

You can say, “I’m not convinced yet.” Ask where the added value comes from, what supports the comparison and what you would give up by changing direction. You are giving the idea something to answer for.

AI can be too ready to agree with you, and it can also keep defending a weak suggestion. Judge the reasons and evidence it gives you. Change the plan when you understand why the alternative is worth choosing.

Link to this answer

It helps to learn how to give a clear brief. Who is the work for? What information does the assistant need? What would make the result useful?

Those are thinking and communication questions. Tools have their particular features, and some are worth learning, but no collection of clever phrases can decide your purpose for you. I would begin there and learn the technical details that actually help with the job.

Link to this answer

Start with what you need to trust it for. A possible plot for a story and a statistic in a report need different kinds of checking.

For an important factual claim, ask two questions: does the source exist, and does it support this particular statement? A perfectly real paper can be used to support something it never said.

Imagine an AI answer claiming that a four-day working week makes every organisation more productive. It links to a report about a six-month trial in a handful of volunteer companies. Even if the report is real and those companies improved, you still need to examine what was measured, how the trial was run and whether the findings apply to your organisation. The word “every” is doing more work than that evidence can support.

Then examine the reasoning and the missing context. Confidence is part of the presentation. You still have to decide whether it has been earned.

Link to this answer

Tell me what you want to do, and I can give you a more useful answer than a league table.

Try a representative task and compare what you can actually use. Include options beyond the names you encounter most often, such as European developer Mistral AI. The field keeps changing; your task gives you a reason to choose.

I worked with both ChatGPT and Claude on the book, sometimes asking one to examine the other’s work. That can produce useful objections, but agreement between two models does not settle the question. Important claims still need checking against evidence.

Link to this answer

A conversation can carry forward its mistakes as faithfully as its useful decisions. If an assistant keeps returning to an assumption you have rejected or repeating a mistake you have already corrected, I would usually start a new conversation.

First, prepare a handover: what you are trying to achieve, what you have decided, which material is current and what remains unresolved. Mark the mistakes and discarded ideas that should stay discarded. You can ask AI to draft the summary, but check it yourself before carrying it forward.

More room for context does not remove the need to choose what belongs in it. One Sage wrote twenty-six pages for his successor while we were working on the book. Most everyday handovers can be considerably shorter. They need to give the next conversation a sound starting point.

Link to this answer

For a substantial project, I keep a separate record outside the chat: the purpose, important decisions and their reasons, the current material and the questions still open. When something changes, I update that record. It is much easier to continue the work when you can show where it stands.

Bring the relevant record into the next conversation and make clear which version is current. Giving the assistant a familiar name does not guarantee that it has the history it needs.

Before consequential work, ask it to explain the goal, constraints and decisions it is working from. Correct any gaps before they become part of the next result. The record helps me keep track of my own thinking too.

Link to this answer

Decide what you are entitled to share before you paste it. Consider who the information concerns, your organisation’s rules and the service you are using, including its terms and settings.

Consider the law that applies where you work as well. In the EU, that includes GDPR obligations when handling personal data and any AI Act requirements relevant to your role and use of the system. These address different obligations; the AI Act does not replace GDPR. For work involving other people’s data, check your organisation’s approved process and ask its privacy or legal lead when you are unsure.

Often you can explain the task with fewer identifying details. Removing a name does not necessarily make the information anonymous. A chatbot’s willingness to accept a file tells you very little about whether you should upload it.

Link to this answer

Start with clear intent and a structure that holds the project together. What are we trying to achieve? How does the material relate to that purpose? Which decisions still stand, and what is open for discussion?

With a tool that supports enough context, your brief can include several books, research reports and earlier drafts. Explain what each source is doing there. The assistant needs to understand how to use the material as well as what it contains.

For a book project, one file might show the author’s voice, another record editorial decisions and a third contain research to examine. Label those roles and identify the current versions. A discarded draft should not quietly overrule a decision you made yesterday.

Ask the assistant to explain its understanding before it proceeds. A useful brief gives the collaboration direction and keeps the next contribution connected to the project you are actually building.

Link to this answer

First ask yourself: what are my strengths as a thinker, and what am I doing to keep them? Perhaps you are good at building an argument, spotting connections or asking the question that changes a discussion. How did you develop those abilities, and where do they get regular practice now?

This is where CompensAItion comes in. Choose which abilities you want to maintain or strengthen, then give them meaningful work. If you delegate an activity, notice whether you are also removing your main opportunity to practise it.

AI can help you arrange that practice. Build your own argument, ask it to challenge the reasoning, then decide what to revise and explain the result yourself. You can use the collaboration to work on a weakness as well as exercise an existing strength.

The aim is to become more capable over time. Check what you can explain and do yourself, as well as the quality of what you produce together, and adjust your practice accordingly.

Link to this answer

CompensAItion is my approach to keeping our thinking abilities in use and developing them while benefiting from AI.

A priority for me is building and maintaining a rich body of knowledge that we understand deeply. Complex thinking needs context. When all we have are a few disconnected facts, there is very little to build with. We need knowledge we can draw on, connections we understand and questions we have spent time exploring ourselves.

The subject can be football, car engines, nail colours or obscure comic books. Take an interest seriously enough to recognise patterns, understand what changes and explain how its parts fit together. Keep asking questions that make you go deeper.

There should be things you can think about in depth without asking a chatbot to supply every connection. AI can help you explore them. Maintaining and expanding your own understanding remains part of the work.

Link to this answer

No. When the deadline is approaching, take the help that lets you get the work done. You do not have to turn every Tuesday into a test of intellectual endurance.

Just remember to compensate when you have room again. Return to something worth understanding. Work through a difficult question, build an explanation yourself or spend time on a subject you want to know better.

Make room for that practice across your life. Some days you need the quickest route to a finished task. On others, you can enjoy taking the long way and finding out what is there.

Link to this answer

Of course. I would actively encourage regular time away from it to reconnect with the world around you. Go for a walk, make something with your hands or have a conversation that you have no intention of turning into content.

Give your attention to people and things that do not arrive as an answer in a chat window. There is plenty to discover there.

You may also choose to work without AI because you want to practise, protect private information or simply enjoy doing something yourself. You do not need to find a technological justification for that choice.

Link to this answer

Our brains evolved to help us survive. They do not automatically put truth ahead of every other concern. Hugo Mercier and Dan Sperber argue that reasoning evolved to help us produce and evaluate arguments in our dealings with other people. That helps explain why we can be so good at defending a belief we already hold.

A belief may also support someone’s identity or place in a group. Challenging it can feel like challenging much more than a claim. Ours included.

Before adding another fact, try to understand what the person believes and what it means to them. Then examine the reasons together. You may notice something the other person has missed. They may do the same for you. Make room for both.

Link to this answer

Begin by describing the other person’s position in a way they recognise. Ask if you have understood it. Identify what you agree about, then make the actual difference clear.

You may still disagree strongly. At least you will be working on the same disagreement, with less need to spend the next ten minutes correcting what each of you allegedly said.

Link to this answer

I wrote it with substantial help from ChatGPT and Claude. They helped with translation, research, drafts, revisions and preparation of the manuscript. The acknowledgements say so openly.

My job included choosing what to keep, questioning what sounded convincing, rejecting suggestions and deciding when a better idea should change the plan. I take responsibility for the finished text, including its errors. The collaboration described in the book was also part of how it was written.

Link to this answer

There is no entrance examination.

I am interested in people willing to examine an answer, including their own, and practise ways of thinking more clearly. You might be a teacher, a manager, a parent, a student or somebody trying to make sense of a difficult decision.

Bring curiosity and a question that matters. We can work on the rest.

Link to this answer

Choose one method and use it on something real. Sketch your thinking before asking for a proposal. Rebuild an explanation with the Pearl Necklace. Decide which part of a task you want to keep practising yourself.

Afterwards, notice what changed. Did you find a missing assumption? Ask a better question? Understand the choice more clearly? Give the method a second outing in a different situation and see what it teaches you there.

Link to this answer

Some examples and descriptions of tools will age. They already changed several times while I was writing.

The book’s foundation is more durable: understanding your purpose, examining evidence, explaining a connection and deciding how to work with someone or something else. Those practices were useful before chatbots arrived.

The Reader’s Companion gives us a place to add exercises and revisit examples as the technology changes.

Link to this answer

They need reasons of their own to learn, and an environment that takes those reasons seriously.

In my Manifesto of a Curious Child, I imagine a learner who has a say in their education, follows questions that matter to them and enjoys stretching beyond what they can already do. They know why they are learning and can turn an idea into something they have made themselves. They also have people to learn with and a community to contribute to.

Difficult feelings and questions belong in that learning too. Children need room to explore who they are, understand conflict and learn to act without harming others. Growing up involves more than producing correct answers.

Teachers, parents, friends and AI can support that growth. A child should be able to recognise what they are gaining from the work, including when it takes effort. The experience of becoming more capable gives learning a value that no machine can simply hand over with a finished assignment.

Link to this answer

A question can open a door. It can also become infuriating when you need somebody to explain the thing you cannot yet see.

In the Manifesto of a Curious Child, the learner tackles challenges just beyond their current abilities, with support close at hand. The useful question for a teacher or tutor is what would help this learner take the next step.

That might be a hint, feedback, a worked example or a direct explanation followed by a fresh attempt. Keep the learner’s purpose and growing understanding in view. An endless sequence of riddles is a teaching style, not a guarantee of learning.

Link to this answer

Begin with why the work is worth doing. If the only visible purpose is to deliver a document and collect a grade, a machine offers an obvious shortcut. Give learners a question to investigate, something to create or a contribution whose value they can understand.

Then make room for the thinking to become visible. Ask a student to explain a choice, work through a new example or show how they assessed the assistant’s suggestion. Let them describe what they tried, what surprised them and what they can now do that they could not do before.

This follows the Manifesto’s view of teachers as allies in learning. We can take AI assistance seriously while still expecting students to understand their work and develop abilities of their own.

Link to this answer

It matters, but responsibility reaches beyond the quality of an answer. In Chapter 3, I use seven rings of responsibility to examine what we do with AI:

• Environment: What resources am I using, and is the use worthwhile?
• Privacy and consent: Whose information am I using, and do I have permission?
• Truthfulness and labelling: What needs to be made clear so people are not misled?
• Attribution and credit: What human work and AI assistance should I acknowledge?
• Social impact: Will this strengthen or crowd out human connection?
• Biases: Who might this exclude or treat unfairly?
• Good will: What am I trying to do, and am I willing to take responsibility for it?

These questions guide my own choices. They do not remove the responsibilities of developers, organisations or public institutions. As the chapter concludes, a ring on one person’s hand cannot keep an entire industry safe. We also need accountable systems and meaningful ways for people to challenge decisions that affect them.

Link to this answer

Because I can see valuable things we can do with it, and I want to help shape the conditions of that collaboration.

Concern can lead us to check a claim, change a workflow, refuse a use or demand stronger oversight. I still want to discover what we can make together. I also want people to retain a say in what is worth making and what kind of future they want to live in.

Link to this answer