Every few weeks a headteacher asks me some version of the same question, usually after a governors' meeting where somebody has said the word "future" with a worried face. What should we be preparing children for now? Behind it sits a fear that the answer has changed and the school has not.
My answer is that the answer has changed less than the panic suggests, and more than most curriculum plans admit. Schools should stop trying to prepare children for jobs that nobody can see and start preparing them for a decision they will make every day for the rest of their lives: what to hand to a machine, and what to keep. Almost everything else a school needs to do about AI follows from taking that one decision seriously.
The Short Answer
Schools should prepare children to know enough to judge, to hand work over to AI well, and to do the things a machine cannot do for them. Each part carries a consequence that leadership teams tend to underestimate.
Knowing enough to judge means subject knowledge matters more, not less, because a child who knows nothing about a topic cannot tell a good answer from a fluent one. Handing over well means children need a habit, practised for years, of deciding which parts of a task are worth doing themselves. Doing what a machine cannot means the school has to protect time for wondering, for caring about other people and for speaking and listening in a room, because those are the things nobody will be able to outsource.
Why Skills Lists Are the Wrong Starting Point
Most answers to this question are lists of skills, and lists of skills do not change what happens in a classroom on a Tuesday. Critical thinking, creativity, adaptability, collaboration, AI literacy: I have yet to meet a school that disagrees with any of them, and I have met very few that could show me where in the week a child gets better at one.
The list is comfortable because it commits the school to nothing. It also rests on a guess about the labour market that no one is in a position to make. When I was in the classroom, and later on a senior leadership team, we were told to prepare children for jobs that did not exist yet. That advice was sincere and useless. You cannot design a curriculum around an absence.
What you can design around is a behaviour that will be true whatever the jobs turn out to be. Every child in your school will spend their adult life alongside systems that produce competent work on demand. What will decide whether they thrive is whether they can look at a piece of work and know which part of it needs to be theirs.
Know Enough to Judge: Why Knowledge Now Matters More
The most counterintuitive thing I tell school leaders is that AI makes a child's own knowledge more valuable, not less, because judgement is impossible without it. The argument that children "can just look it up" was weak before generative AI and it is wrong now. A model will produce a confident, well-structured answer to anything, and the only person who can tell whether it is any good is somebody who already understands the subject.
The Curriculum and Assessment Review, chaired by Professor Becky Francis and published by the Department for Education on 5 November 2025, reached the same conclusion from a very different direction. Its final report says that "subject-specific knowledge remains the best investment in preparing young people for these challenges and opportunities", and that young people need to "learn how to use AI effectively, without becoming dependent on it". It also dismantles a comfortable assumption: "the assumption that young people will acquire digital literacy automatically is incorrect."
Read those three sentences together and the leadership implication is clear. A school that thins its knowledge base to make room for "AI skills" is removing the one thing that makes those skills usable. Outsource the doing, not the thinking, has been my governing principle for years. You cannot follow it if there is no thinking to keep.
Hand Over Well: The Habit That Decides Everything
The second preparation is a habit, not a lesson: children need years of practice at deciding which parts of a task to give to a machine and which parts to keep, and they need adults who make that decision visible. Most schools currently have nothing here. AI use is either banned, so children learn the habit at home with no guidance, or permitted, so they learn whatever the tool teaches them.
We now have a clear picture of what the tool teaches them when nobody else does. A study by David Strömberg, Victor Lei and Yanhui Wu, published as a CEPR discussion paper on 2 June 2026, followed 26,811 secondary students in China over thirty months. Students who adopted generative AI saw homework scores rise by 18 per cent and completion time fall by 30 per cent. Their monthly exam scores fell by 20 per cent within six months, and the effect on high-stakes entrance exams was larger still. About 80 per cent of the AI users showed the pattern the authors describe as homework outsourcing: exceptionally fast completion paired with high homework marks. The students who kept working at a normal pace, using the tool alongside their own effort, showed minimal learning loss.
That is the handover decision, measured at scale. Same tool, same children, two completely different outcomes, and the only thing separating them was what they chose to keep. A school that does not teach that choice is leaving the most consequential learning habit of a child's life to chance.
The evidence that children are already making the choice, unsupervised, is not in doubt. Pew Research Center's report How Teens Use and View AI, published on 24 February 2026, found that 54 per cent of American teenagers had used chatbots for help with schoolwork, and that one in ten did all or most of their schoolwork with them. I have not met a British school leader who believes the picture here is different.
Do What a Machine Cannot: Wonder, Care and Presence
The third preparation is to protect the parts of childhood that produce the things a machine cannot: wondering about questions nobody set, caring about people and consequences, and being present with other human beings in a room. Machines can compute. They cannot wonder. They cannot care. Those are not soft skills to squeeze into a Friday afternoon. They are the reason a human being will still be in the loop in twenty years.
This is also where I part company with the fashionable idea that the future belongs to "AI-native" children who grew up with the tools. Fluency with a tool is the easiest skill there is to acquire. The OECD has recognised as much: its draft framework for PISA 2029, released in February 2026, adds Media and Artificial Intelligence Literacy as an assessed domain and describes the aim as students who can "make informed and ethical decisions about when and how to use digital and AI tools". Note the wording. Not how to operate them. When and whether to.
In school terms, this means oracy is not a bolt-on, group work is not a soft option, and a child spending an afternoon on a question with no mark scheme is not a waste of curriculum time. It is the curriculum doing its job.
What I See in Practice
Across the schools and trusts I work with, the strongest approaches to this question share one feature: the leadership team has decided what children should keep, and every teacher can say what that decision is. The weakest have a policy about AI and no view about learning.
The pattern usually appears in the first hour of a strategy session. I ask the senior team to name a task children are set every week and to tell me what a child would lose if a machine did it for them. In a strong school the answer comes fast and it is specific: the struggle with the first draft, the working out, the moment of choosing an argument. In a school that is not yet ready, the room goes quiet, or the answer is "the grade", which is not a loss to the child at all.
This is where the conversation becomes difficult, because the honest audit often finds that a good portion of what children are asked to do would lose nothing by being handed over. That is uncomfortable, and it is also the opportunity, if the school uses the time for the things only a human can do.
I taught in secondary classrooms and sat on a senior leadership team before I wrote books about AI in education, and the thing I remember most clearly about those roles is how rarely we had the time to ask why a task existed. AI is forcing the question, which is the most useful thing it has done for schools so far.
The Handover Question
The Handover Question is the one test I ask school leaders to put to every task children are set: if a machine did this for them, what would they lose? If the answer is nothing, hand it over and use the time better. If the answer is the learning itself, protect it, and be able to say why.
It works because it is small. It needs a department meeting and some honesty, not a new curriculum. Run it across a scheme of work and three kinds of task appear. Tasks that were only ever about producing a thing, which AI can now produce: reclaim the time. Tasks where the making was the learning: protect them, and tell children why they are being asked to do something a machine could do faster. And a third kind, which is the interesting one: tasks that could be redesigned so that the child uses AI for the doing and is assessed on the thinking, through the process they followed and how they explain and defend the result in person, not just the product they hand in.
That third category is where the preparation for adult life actually happens, and it is the one most schools have not started.
The Mistakes I See Most Often
The mistakes I see most often are the ones that feel responsible. The first is banning the tools and calling that a strategy. It protects the school and abandons the child, who learns the handover habit anyway, at home, from a system that has no interest in their education. The second is the opposite: buying an AI platform and treating access as preparation. The Strömberg study is the clearest evidence yet that access without the habit is worse than nothing.
The third mistake is rewriting the skills list and stopping. The fourth is leaving the whole question to the computing department, when the handover decision arrives first in English, history and science. The fifth is trusting AI detection tools to hold the line. They do not work reliably, they punish the honest, and they are a way of avoiding the decision rather than making it.
What Good Looks Like
A school that is preparing children well for this can show you three things: teachers who can say what children must keep and why, children who can explain when they used AI and what they did themselves, and time in the week that is visibly protected for the human work. None of those requires new technology. All of them require a leadership team with a view.
You would see subject knowledge treated as the foundation of judgement, not as a relic of a pre-AI curriculum. You would see the handover taught explicitly from the first year a child touches a tool, with the reasons made visible. You would see assessment shifting so that process and live performance carry weight alongside the product. And you would see a leadership team that could answer the first question of the Readiness Test, What is AI for here?, with a sentence about learning rather than a sentence about risk.
What to Do Next
Start by running the Handover Question across one department's schemes of work before you write another line of AI policy. It takes one meeting, it produces the three categories above, and it gives the leadership team something concrete to decide. Then decide it, write it down in a form staff can act on, and say the same thing to parents, because they are asking the same question you are.
If your school's AI document currently tells people what they may not do and nothing else, my article on the difference between an AI strategy and an AI policy is the next thing to read. If you want to know whether the school is ready to hold these decisions consistently, what it actually means to be AI ready sets out the test I use.
I work with schools and trusts to turn this question into a shared position on what children should keep, a set of priorities for the curriculum, and a plan the whole staff can follow. If your leadership team is working through it now, that is the kind of work I support through school AI strategy sessions and leadership advisory.
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and a bestselling author on AI in education. He was a secondary teacher and assistant headteacher before he began advising schools, trusts and governments on how to lead through AI.
