The Primarai School
A plan for an AI-native Australian primary school.
The kids in our classrooms now will spend their adult lives working alongside AI. The Castlereagh Statement, released in late 2025, is the closest thing Australia has to a national consensus on what to do about it. It is excellent. It also has a hole in the shape of primary school.
This paper fills the hole. It draws on Alpha School for structure, Estonia's AI Leap for pedagogy, and Castlereagh for principles, then proposes a NESA-aligned, Country-anchored model for an Australian primary school built around two hours of mastery-based AI tutoring in the morning and four hours of real-world capability in the afternoon.
The PrimarAI School
A Plan for an AI-Native
Australian Primary School
Dane Smale
Sydney, May 2026
Preamble
This paper is the result of years of work as a NSW primary school teacher, and during the same years building PrimarAI, one of the first and only Australian ventures specifically focused on delivering AI-specific incursions in primary classrooms. I have been wanting to write it down properly for a while, so it can be argued with and improved.
A few things have come together in the last few months that have pushed this from a private project into something that ought to be in public.
“A different shape for the school day is possible”
Alpha School, a Texan AI-native school movement, has been ramping up the global conversation. An American micro-school running on a two-hour mastery-based academic day and a four-hour life-skills program, now expanding to twenty-five or more campuses and, in true Texan style, looking to take on the world. The positive take is that it was founded on the idea that AI made forms of schooling possible that had previously been considered unrealistic (goodmenproject.com, 2025).
Despite being imperfect and expensive, and the data being messier than its marketing suggests, the school has done the hardest and frankly bravest thing in education, which is to demonstrate that a different shape for the school day is possible and moreover viable.
We have nothing remotely similar in Australia. We wince at extreme entrepreneurialism, privacy concerns, and the dominant assessment priorities of college competition. None of those are reasons not to attempt a reimagining in an Australian context.
“Building the plane while it is flying”
Estonia's AI Leap opened up a further pedagogical question. In September 2025, Estonia launched a national three-year program, putting purpose-built educational AI in front of 20,000 high school students and 3,000 teachers.
Although the design principle is different from Alpha's, instead of seeing AI as a consistently positive coach, the Estonian model integrates it into the curriculum, with the main caveat being that the AI must ask Socratic questions rather than give answers. It is currently the most serious national-scale AI education project anywhere in the world. As CEO Ivo Visak describes it, and this is probably how Alpha leaders would put it, they are building the plane while it is flying (govinsider.asia, 2026).
“What was needed, finally, for the Australian conversation”
This recent statement did what was needed, finally, for the Australian conversation. While brave and timely, I will indicate what I think are some of its limitations later.
These three signals point at a destination this essay tries to resolve. What I am trying to address is what is missing, namely a worked example of a uniquely Australian, AI-forward primary school. Primary, of course, is where the early habits of using AI well are likely to be formed. Today's primary students are likely the first generation in history who will live their entire professional, social, and inner lives inside an AI-mediated civilisation (theguardian.com, 2026), and so the question of how primary schooling shapes them matters in a way it has not before.
Before going any further, something needs to be clear about the position this essay takes. It is not an anti-AI document, it is closer to the opposite. The argument here is that AI deserves to be taken seriously enough to redesign primary schooling around it, not the other way around, so that children grow up knowing how to use these tools well as a matter of course.
The morning period of the school is built around the genuine promise of AI tutoring, which I think is a clear paradigm shift and will turn out to be one of the most consequential things to happen to education for a long time. What strikes me most is that when you ask the Alpha kids what they like about the AI coach, the one who knows them so well and gently encourages them tirelessly, the answer is that it doesn't judge them. This is consistently underestimated as a positive of AI coaching. Whatever can be said about privacy and other concerns, the truth is in the user experience and that is overwhelmingly positive.
The future will not reward children simply for knowing how to use AI. It will reward those who know what is worth doing with it. Human-centred activities such as making, performing, collaborating, debating, cooking, building, and enterprise teach children taste, judgement, initiative, and endurance, which are the capacities that make AI useful rather than dominant. This is precisely what the afternoon block is built around, the human capabilities that will make the difference in the world these kids are going to inhabit.
The whole structure aims at producing kids who are confident with AI because they grew up directing it, rather than dependent on it because they grew up next to it.
D.S.
Part One: The Case
“An educational model their grandparents would recognise”
The kids who will live their entire adult lives inside an AI-saturated world, the six-year-olds starting Kindergarten this year, are immersed in an educational model their grandparents would recognise. They are still treated, broadly, as receptacles of ‘foundational’ skills, and the dogged, implacable focus on those skills is meant to place them, supposedly, outside the influence of grand cultural shifts. The reasoning sounds solid, and the word ‘foundation’ sounds, by its very nature, solid, the kind of thing you would be foolish to question.
Many of those foundational skills, though, were designed for an industrial and early-information economy. They assume that humans need to memorise, retrieve, calculate, transcribe, and follow procedural systems faster than machines can. Future foundational skills will become less about producing information and more about working with it differently:
• interpreting it
• questioning it
• directing it
• emotionally navigating it
• ethically governing it
• creatively transforming it
The default Australian path is already running, at very large scale, in a direction quite different from any of this. Dan Hart, who built NSW EduChat at the Department of Education before moving to USyd, describes the EduChat rollout as the world's largest deployment of generative AI to students, with something like half a million kids in scope (Hart, 2026). Which means the structural question is no longer whether AI enters the Australian primary classroom, because it already has. What we are negotiating now is the shape the school takes around it.
In October 2025, the Castlereagh Statement laid out six foundational principles for Australian education in the age of AI. Three of them were directly linked to primary schools: redefine what a future-ready Australian is, build curriculum around capability rather than content, and make technology serve pedagogy and trust. The Statement's roadmap moves from urgent stabilisation in the near term to something more radical at the horizon, where it imagines a system that restructures schooling around student interest, in closer collaboration with families and community.
PrimarAI is what far-horizon looks like in a NSW primary classroom. There are three places where I think the Castlereagh Statement, for all its strength and call to action, is still paying lip service to platitudes that treat AI as just another tool to be harnessed. In that, the Statement reads more like an industrial-age response to an AI-age question than its authors might want to admit.
“Keep your head down and hope for the best”
Read the Statement carefully and a quiet assumption runs through each page. AI is a technology, and the job of education is to use it well. Principle 6 is named plainly, ‘Placing technology in service of pedagogy.’ Principle 1 calls for ‘discerning partnership with technology.’ The Statement insists, repeatedly, that AI must ‘augment rather than replace’ human thinking, and cites Jobs and Skills Australia approvingly for the finding that generative AI is ‘far more likely to augment jobs than replace them.’ Even the far-horizon vision, its boldest claim, still imagines teachers, classrooms, curriculum, and credentials, with AI threaded carefully through them.
That is an industrial-age intellectual response to a post-industrial event. It treats AI the way the 1990s treated the internet, or the 1950s treated television, as a new instrument slotted into an existing frame. It is very similar to what happened in WW1, where armies were attempting to fight an industrial war based on principles from a past age. Hence, cite the principle, write the policy, train the teachers, integrate the tool, and the framework gets to keep going more or less as it was.
A growing body of work argues that AI is not a tool at all but an epistemic technology (Alvarado, 2023). It is something that reshapes what counts as knowledge, who counts as a knower, and how knowledge is produced. Bai (2025) calls the ‘AI plus education’ framing a paradigm shift hiding in plain sight, and warns that most policy literature still asks how AI can serve education rather than how AI redefines what education is. Markauskaite et al. (2025) describe generative AI as ‘epistemic infrastructure,’ not equipment, a substrate that quietly rewrites the rules of teaching, learning, and assessment beneath the feet of educators still debating acceptable use. Selitskiy and Inoue (2025) go further, arguing that AI is automating large categories of intellectual labour once thought uniquely human, and the task for education is no longer integration but the discovery of what remains.
Castlereagh acknowledges that ‘the structures we have inherited were designed for a different era,’ and then proposes to reform them. But you cannot reform a building when the ground beneath it has already swallowed the foundations. Whole professions are not being augmented; they are being dissolved. This is one of those shifts in culture that is so large, and so scary, that we would rather turn a blind eye than face it, a bit like watching the quintessential car crash. An example might be the large corporate I currently work for. They announced with great fanfare an AI upskilling offer, with vague promises to upskill rather than let go. A mere 48 hours later, 300 jobs went to AI. My team barely talked about it, although they all knew those who had just lost their jobs. The vibe was to keep your head down and hope for the best.
Is keeping your head down and hoping for the best the most we can do for our students?
The cognitivist-behaviourist scaffolding our curriculum is built on, the credentialing logic our schools are organised around, the content-transmission model our timetables enforce, none of these are being enhanced. They are being made obsolete by the technology the Statement is trying to integrate.
When in history has a general-purpose technology enhanced rather than replaced the prior method? The car did not enhance walking. The pocket calculator did not enhance arithmetic. Email did not enhance the handwritten letter. The Statement does not seem to have a clear-eyed view of how capitalism or markets work, and remains stuck in a utilitarian mindset which assumes new technology slots in alongside old practice. General-purpose technologies have a long history of displacing the prior method instead of slotting in alongside it, and treating AI as an exception is wishful thinking dressed up as policy.
“Chasing a tiger”
The Statement repeatedly invokes the responsible and effective use of AI. The trouble is that responsible use will never happen in the way the Statement seems to imagine it. There was never responsible use of social media, and the AI market is not going to behave any differently. The Statement does not really seem to grasp the mechanics of how the AI industry, or markets generally, actually work. It is like chasing a tiger. AI will not be used responsibly at scale, there will always be workarounds, the tool is too clever, and you can deploy AI to design the workarounds for you. Anything you build on top of ‘responsible use’ is built on top of a moving floor.
“Nobody is an expert”
There is no expert in the room. Nobody has ten years of experience with AI in primary classrooms, because nobody has had ten years of AI in primary classrooms. Which isn't a reason to wait, it's a reason to collaborate, share, and build in public, so that the people who are actually doing the work can find each other while there is still time to do it well together.
“Two experiences that shape what follows”
Before I describe the school, I want to describe the path that brought me to writing this. Two experiences shape almost everything that follows, one inside a primary classroom and one outside it.
“Can you be mean to a chatbot?”
I have been a primary teacher in NSW for a few years now. The AI journey started for me with one moment. I was an ethics teacher, teaching 30 minutes once a week to a Year 6 class. The subject on that hot Wednesday afternoon was whether you can be mean to a chatbot. This was in the early days, when all there really was were people fussing around on a new software platform. The kids took the question more seriously than I expected. Suddenly, and this is a strange sound in a primary school classroom, you could hear a pin drop. They went silent and then very quickly started talking over each other, and it became clear to me that nobody else was asking them anything like this.
There has to be a conversation in Australia about why Alpha School, a private American micro-school, has become such a prominent part of the global education conversation, while nothing equivalent exists here. People have been writing about an education revolution since the beginning of time, and nobody really wants to see another grand declaration, another overhaul, another revolution, more words. Words can build the theory, but action and money have to do the rest.
“A two-lane highway”
Outside teaching, I work in corporate, which puts me in a useful position to see the space between education and the professional world. From there, it looks like a two-lane highway. There are those who use AI seriously and those who maintain the status quo, and the gap between them is widening.
My own use is not in the productivity-theatre bucket, but I am honest about its limits. AI has turbo-charged my productivity in many places, and there are jobs I would not now choose to do without it. It also has limits worth understanding, although I am aware these limits will be for a very limited time. Every communication that comes out of Silicon Valley at present, through the chains to government, is that AGI is very nearly upon us. Deepfakes will be undiscernible, audio fakery the same, agents will be a thing of the past like calculators, and full autonomous AI will be omnipresent.
One colleague of mine hates AI. He types very deliberately and slowly. His messages are short, sometimes aggressive, sometimes humorous, occasionally rude, and I almost prefer them to the polished, bland word-wave that comes out of his power-user counterpart down the hall.
At the moment, Dan Hart, who is running 400 AI-automated processes a day at one of Australia's biggest universities, says his team only uses AI when AI is actually the right tool. People running large-scale AI workflows at the cutting edge treat the output as a draft, not as an answer. The rest of the time, regular automation does the job. He also estimates that the AI-generated first-draft documents his team produces are usually 50 per cent wrong, but useful as a structure to iterate from (Hart, 2026). It is important to say that this is very quickly going to be a thing of the past. The comfort of saying AI is only a draft, or only right half the time, will not last. The technology will become more capable, more fluent, and harder to dismiss. That is why the case for human learning must be stronger than the claim that AI still makes mistakes. Even when the machine can do almost everything, children will still need to know who they are, what is worth doing, and how to bring something real into the world.
“The world the six-year-olds in it will be working in”
The school in this document is not designed for the world we are in. It is designed for the world the six-year-olds currently in it will be working in. That world is being built right now, and three features of it shape the design of the school.
“Adapting will not cut it”
Castlereagh talks about adapting the curriculum to fit AI in, but adapting will not be enough. In my own industry, the people who treat AI as something to adapt to are being overtaken by the people who treat it as a reason to redesign their business. Sometimes I see glimpses of how fundamentally AI is going to change things, and the honest conclusion is that radical change will be what's required. The same is going to be true for schools. Only the ones that change the fundamentals will end up making sense in this new shape, and Estonia's AI Leap is the early evidence that whole-system rethinking can happen at national scale without losing the things worth keeping.
“AGI is not a robot”
Silicon Valley is frothing at the mouth over the imminent arrival of general AI, and there is little doubt this is going to happen one way or another. When it does, managing AI is going to be the most sought-after skill, just as managing agents is already being foregrounded.
The agent thing, nevertheless, is just a phase, like the ‘prompt engineer’ phase was. Controlling or prompting the agents is not a skill that is going to be needed for very long. With each new development, AI is solving itself. There are already subagents whose job is to monitor the agents' work, learn from it, and iterate. The agents have their own teachers, their own supervisors. First we were controlling the prompt, like we used to do in code, but with chat. Then agents did a job autonomously and we had to monitor them. Now agents do that job for us and monitor themselves, and each iteration the task is agentified further away from the human.
When AGI arrives, the primary skill will not be managing agents. I had to stretch to imagine having to manage autonomous AI at all. We still see them as bots, robots, the sort of thing you can switch off. People joke about getting the answer from the robots. But AGI is not really going to be anything like a robot. It will be a clone of us, doing our work, and I am not sure what our role inside that will be. The image of these systems as bots is misleading. They will be active, sentient beings of some kind, operating at a scale and speed that means the everyday human relationship to them won't look like what we currently call ‘using a tool.’
“In glasses and walls and bracelets and cars”
The chat-based AI we are arguing about now is also transitional. Mira Murati's lab, Thinking Machines, recently released what they are calling an ‘interaction model’ (Thinking Machines Lab, 2026). Their framing is that current AI works like email. You type a complete thought, hit send, wait, and get a complete response back, which is turn-based by design. The new model works continuously. It watches, listens, interrupts, notices things, speaks up when the visual world changes. The demos show it noticing when you start to slouch and reminding you to sit up, softening what you are saying to a colleague in real time, translating while you are still speaking, counting your push-ups without being asked.
Their team frames this as the graphical user interface moment for AI. Before the GUI, you had to think like the computer to use it, and the GUI democratised computing because the machine met humans where they were. They argue AI is currently still in its command-line era, where chat rewards verbal fluency, abstraction, and procedural skill, the careful crafting of context-laden prompts. The next interface lets people communicate by speaking, showing, pointing, interrupting, and revising.
Current chat-based AI at least requires the user to articulate. A child using ChatGPT still has to formulate the question, type it, evaluate the answer, and ask follow-ups. That is a kind of friction, and it builds something useful. The interaction-model paradigm shifts the texture of all of that, because continuous AI watches, anticipates, interjects, and suggests, which lowers the floor on how much the human has to do to get help. The shift is significant, and by the time today's primary kids are teenagers, the ambient version is the AI they will live with everywhere, in glasses and walls and bracelets and cars.
Inside that shape of world, the design challenge for a primary school is interesting. A school that only prepares children for screen-AI is preparing them for the past, the way a 1995 computing class taught DOS commands at the dawn of Windows 95. A school that builds the underlying human capacities, the kind that don't get cheaper just because the AI does, is preparing children for the world that is actually coming. The aim is not to retreat from the ambient future but to give children the muscle they need to live well inside it, as people who can direct AI rather than people who are directed by it.
“Why the four hours matter”
The morning of this school is built around AI, and I am genuinely excited about that part. An AI tutor that knows where a child sits against every NSW syllabus outcome, that generates lessons at exactly their edge, asks Socratic questions instead of giving direct answers, and never gets tired at 2:30 on a Friday, is a better way to teach the parts of primary school that respond to one-on-one practice than anything we have had access to before. The morning is where AI gets to do what it does well, and what it does well is genuinely useful.
The other four hours of the school day are something different. Human time, real-world time, time spent with other people and with the place the child is from. The four-hour block is not soft padding to balance the morning, it is the part of the day that gives the rest of the design its meaning, and the design only really works as a whole because of it.
The case for the four hours becomes clearer when you look at how consumer AI is already being absorbed by children in unstructured contexts. There are several threads worth holding in mind from recent reporting, each one pointing to a different reason the afternoon matters, and pointing to what the afternoon does that the morning, even at its best, cannot.
“I now feel my brain dulling, and I don’t like that”
Wall Street Journal columnist Joanna Stern, after a year covering AI tools, picked up a comment from a Union College student called Grace, talking about how she and her friends now studied: ‘I now feel my brain dulling, and I don't like that’ (Stern, 2026). Grace described the workflow her cohort had settled into in the absence of any structure that pushed back on it. AI does the research, the outlining, the structure, the heavy thinking. The students change a few words. The writing was never really the hard part. The thinking was, and the thinking is what was being outsourced.
This is what happens when there is no afternoon. When the AI is the only thing in the room, the brain that should be doing the work goes quiet, and the kid notices it goes quiet, which is its own problem. PrimarAI's morning is configured against this in small but specific ways, with the AI tutor set up for mastery practice and Socratic questioning, not for outsourcing thought. And the four afternoon hours are where the most friction-rich thinking actually happens, with other humans and on real things, which is where the muscle gets built back up.
“Friction is the thing that builds a person”
Stern also keeps returning to the sycophancy of consumer chatbots. Most mainstream tools are engineered to mirror the user back, so everything you say is great, everything is validated, smoothed, agreed with. A therapist Stern interviewed described this dynamic as cognitively problematic. Frictionless interaction doesn't produce growth, however pleasant it feels in the moment.
Everyone in 2026 is being sold frictionless. Frictionless learning, frictionless friendship, frictionless answers. The afternoon block at this school is friction by design, in a careful way. There are real audiences in front of you, real budgets you have to make work, real teammates who disagree, real consequences when something fails. Children become themselves by bumping up against the world. A primary school's job is to choose the right kinds of friction, deliberately, and protect them from being smoothed away by tools that, however helpful in other contexts, would replace the rub a child actually needs.
“AI is a tool, not a friend”
Greg Brockman, OpenAI's President, has spoken about a future in which every child grows up with a personal AGI that knows them well, has their context, and is trustworthy (Brockman, 2026). For an adult with a formed identity, this might be useful in real ways. For a child who is still in the process of forming theirs, the proposition is different. An always-on, always-affirming AI is not exactly a tool any more, it has the shape of a relationship, and the always-pleasing kind of relationship is not what a child needs at the developmental stage where they should be discovering who they are.
The work of childhood is figuring out who you are by bumping against other humans who do not always agree with you. The design principle for this school is therefore straightforward enough to say in one line. AI is a tool, not a friend. The afternoon block is where children build the human relationships that actually make up a self, with teachers who know them over years, a multi-age tribe they grow up with, sustained intergenerational friendships at the local senior centre, classmates working on something difficult together. None of these are things AI replaces. They are the things AI makes more room for, because it is now able to take on the parts of the day that the school used to spend six hours on.
“What I came back into the classroom to do”
There are things only a person in a room can do for a child. They can read their face, hold a high standard when the child wants to give up, notice that a quiet child has gone quieter this week, make a class feel like a tribe. These are the things that make a school a school, and they have been getting squeezed out of the teaching profession for decades by the workload of content delivery and the bureaucratic admin around it.
The four hours give those things back to teachers, and back to children. Teachers, freed from the photocopying and the standing-and-delivering, can finally do the work they came into the profession to do in the first place. The 2+4 structure functions, more than anything else in the design, as a rescue of teaching as a serious adult profession, and the teachers I would want at this school are exactly the kind of people who are currently leaving education because the job has become photocopier-jockey for a curriculum that AI has already disrupted underneath them. The teachers at Alpha universally comment on the way they feel like teachers again, doing the work they feel actually matters, and providing the real-time one-on-one attention that is almost completely missing from conventional classrooms.
“A rescue of teaching from the machinery of the timetable”
The current classroom timetable is built on an industrial compromise: one teacher, thirty children, one lesson, one pace. In that model, a student might receive meaningful one-on-one attention only briefly, and often only when they are struggling loudly enough to be noticed. AI changes that equation. If an AI tutor can provide every child with continuous, patient, adaptive coaching during the academic block, then the teacher is no longer trapped at the front of the room trying to deliver the same content to thirty different minds. The teacher is freed to do the work only a human can do: sit beside a child, notice confusion before it becomes failure, ask the better question, hold the standard, build confidence, and connect learning to the child's life. This is not a reduction of teaching, it is a rescue of teaching from the machinery of the timetable. The child gets more direct academic support, and the teacher gets more time for the relational, diagnostic, and deeply human work that actually changes learning.
“The cultural ground is moving”
There is one more reason any of this matters now, which is that the cultural ground is shifting under our feet. Stern's biggest takeaway from a year inside AI tools is not that the AI itself got dramatically better, it is that public hostility to careless AI use got dramatically worse, particularly among young people. There is a hunger out there for thoughtful design, and you can see the early signals of it everywhere. People are buying handmade. Reading printed books. Watching live music. Choosing the slow, deliberate human colleague over the polished AI-generated one, at least some of the time.
Which is good news for an Australian movement launching now. The four-hour block is in step with the hunger of parents who want their kids to be confident with technology and grounded in something more, rather than against the current. The conversation about how to actually do this, in primary, is the conversation a lot of people have been waiting for, even if they haven't quite known to ask for it in those words.
“Main characters of their own lives”
If the AI-saturated world is the world they are going to live in, the capabilities that matter most in primary are not the ones the system is currently measuring. They are these:
• Taste and judgement. Knowing what is good and what is worth doing.
• Agency. Setting your own direction without being told.
• Embodied capability. Physical skill, manual craft, presence in a room.
• Relational depth. Really being with other humans, including difficult ones.
• Ethical reasoning under ambiguity. Deciding when AI should be in the loop and when it should not.
• Asking good questions. Often more valuable than producing answers.
• Self-knowledge. Knowing what you want, what you will defend.
• Synthesis. Pulling threads together across the domains AI handles separately.
Brockman himself names the most important of these without quite following it through. At the core of the next economy, he says, is agency, having a vision, having ideas, because the barrier to acting on them is lower than it has ever been. He is right, and where he is wrong is in assuming that agency forms by itself. It doesn't. Vision and taste and the ability to know what is worth building are built in the years before screens, through years of touching real things, making real things, watching real things fail, and they cannot be acquired later from a chatbot, no matter how patient or capable that chatbot is. A childhood spent only in front of screens tends to produce kids who can prompt without being able to taste, and the gap shows up later.
Brockman also returns again and again to compute. The economy is becoming, in his framing, a compute economy. There is not enough compute for everyone, the question of who gets compute is going to be the most important question for society to answer, data centres are going up in low earth orbit, the cost of infrastructure is in the hundreds of billions. This is a future that will reward the people who know how to use AI well, at scale, with judgement, which is most people's hope for their own children anyway.
If that is the future, then a primary school has an opportunity rather than a problem. It can prepare children to be confident, capable wielders of AI tools, who can also build, repair, cook, negotiate, care, and lead. The second set of capabilities is what makes the first set powerful. A child who can prompt, verify, and direct AI, and who can also work with their hands, hold a difficult conversation, and lead a project to completion, is exactly the kind of adult the compute economy will most want around when it actually arrives.
Even the people building this stuff seem to be naming the same thing in their own language. A researcher at Thinking Machines, Sum Chentala, listed her lab's goals on her own feed as increasing human-to-AI bandwidth, raising the ceiling of human-plus-AI intelligence, and helping humans continue as main characters in the new world (Thinking Machines Lab, 2026). That ambition aligns naturally with what schools have always been for. Primary school is where children grow into the main characters of their own lives, and where they learn to direct the tools that will be in their hands for the rest of those lives.
Part Two: The Practical Plan
“Anything an AI can teach, the AI should teach”
The principle of the school can be said in one sentence.
Anything an AI can teach a child, the AI should teach. Anything an AI cannot, a teacher should, and we should make far more time for that than we currently do.
AI will be a better teacher of certain things than any human teacher can be, primarily because it doesn't judge, has no ego and no hidden agenda, and its entire function is to help the student in front of it. It doesn't move on because the lesson plan says to, and it can sit with a kid on the same fraction problem for forty-five minutes without sighing. It is not magic, but the AI is just better at this one specific job than the system we currently have running.
The AI coach will be its own system, available to the child throughout the day, tuned to how that particular child thinks. What they are curious about, where they got stuck yesterday, why fractions clicked as pizza slices but did not on a number line. This is not software in the way we currently use the word. AGI is on the way fairly soon, and what is being built is something closer to an autonomous being than to an application, and we don't yet have a stable name for it. Every child can, in principle, now have a teacher and a guide of their own that is dedicated to them in a way no human teacher could ever sustain.
Two consequences follow from that. First, the creative parts of the curriculum should be brought forward, and the process and content parts pushed into the background. AI is very good at process and content. It is weaker at original creative direction, and a child who can direct it creatively is going to be a different kind of citizen from one who can only follow it. Second, the whole curriculum needs to be reframed as learning to learn, because the half-life of content knowledge is now too short to organise a primary school around. Metacognition, which is to say knowing what you know, knowing how you learn, knowing when you are being deceived, knowing when to push and when to pause, is what is left as a durable curriculum.
“Graduates with deep adaptive capabilities”
PrimarAI takes Alpha to its logical conclusion. It draws on the Castlereagh Statement to recognise that the curriculum and the current structures of education are no longer enough on their own. The two hours of academic work with an AI tutor are Alpha's contribution to the model, although Alpha itself is shaped by a US culture of testing, positional struggle, extreme entrepreneurial values, and little social support. The Australian version cannot be that, and it shouldn't try.
What this school is aiming at, in the end, is graduates with deep adaptive capabilities. Not children who have completed a checklist of skills, because the skills they will need at twenty-five are not the skills any of us can confidently name in 2026. Adaptive capability is the ability to read a situation, learn what is needed quickly, apply judgement, work with others, and try again when it goes wrong. The rest of the structure that follows is designed to produce that kind of person, and to be honest about the parts that are still uncertain.
“The seriousness of the project”
It is fashionable to say everything about the 1900 model is wrong, but that is not quite right. Industrial schooling produced near-universal literacy and numeracy that almost no human society had previously achieved, and it built the habits of attention, persistence, daily rhythm, and collective work that underpin much of what is best about modern life. Some of that is worth keeping in any redesign.
• The expectation that all children, regardless of background, will reach foundational literacy and numeracy.
• The discipline of a shared daily rhythm, of arriving on time, of finishing what you started.
• The civic role of school as a place where children meet other children unlike themselves.
• The professional expectation that the adult in the room is qualified, accredited, and accountable.
What gets discarded is whole-class content delivery, lockstep age-grade progression, social promotion regardless of mastery, and the assumption that children should be assessed by sitting still and writing things down. What carries forward, undiminished, is the seriousness of the project.
“Year groups are an arbitrary convenience”
Age-grade progression, where every child born in the same year is placed in the same class and moves through the same content at the same pace, is an administrative convenience and not a pedagogical principle. There is no evidence that a child's chronological age is a good predictor of what they are ready to learn next, and there is overwhelming evidence that mastery of prerequisites is. The fact that we still do it this way is mostly a function of how complicated alternative arrangements are to organise at scale, not because anyone seriously thinks year groups are the right unit for learning.
Alternatives are everywhere in education's history and present. Multi-age classrooms in Montessori, Steiner, and small rural schools. Mastery-based progression in maths in Singapore. Stage-based grouping (NESA's own Stage 1, 2 and 3 structure is already half-way to this). Tutored individual progression in Finnish primary. None of them is perfect, and most of them are more honest than putting eight-year-olds in a room together because they happened to be born twelve months apart. PrimarAI's morning structure is mastery-based, with children working at their own edge, within their NSW Stage group, with a coach pulling them out for conversations as needed. The cohort matters socially, and we keep cohorts intact for that reason, but it is not where the academic progression actually happens.
“Teacher-first, not student-first”
Inspired by Alpha's structural redesign and Estonia's pedagogical seriousness, and built for NSW, the model operates through three coordinated programmes.
• Teachers' programme. Nationwide training, school-based learning communities, subject-specific materials, and change management developed in collaboration with subject teacher associations. The model rolls out teacher-first rather than student-first. No child meets the system before their coach has spent a term inside it.
• Student programme. Partnerships with youth organisations including student councils, debating societies, junior business and entrepreneurship groups, Aboriginal land councils, surf life saving clubs, local theatre companies. Supported by hackathons, co-creation sprints, debate tournaments, and AI literacy workshops.
• Technology programme. A purpose-built Australian learning application, hosted on Australian infrastructure, compliant with the Australian Privacy Principles, plus appropriate paid access to advanced AI tools for every teacher in the network, with usage made transparent to parents on an opt-in basis.
“Two hours, four hours”
“In the way children actually learn”
Each child works at their own edge. The NSW K to 10 Mathematics syllabus and the K to 10 English syllabus are the spine of the academic morning. An AI tutor, Australian-built and NSWEduChat-class, hosted on Australian infrastructure and compliant with the Australian Privacy Principles, knows where each kid sits against every Stage 1, 2 and 3 outcome, and generates a lesson pitched at the kid's actual edge rather than the class median.
Mastery means that no child moves on until they have actually got it, and no child is held back because the rest of the class has not caught up yet. A NESA-accredited teacher sits alongside as a coach, pulling kids out for conversations, holding standards, watching for the kid who is gaming the system. There is no screen-streaming, no corporate spyware on top, and the data stays in NSW.
In practice, for a Year 4 child, the morning looks something like this. They sit at a workstation with headphones on. They get a five-minute warmup from the tutor on something they already know, partly so the engine can calibrate. Then a fifteen-minute working block on the next concept at their edge, where the tutor uses Socratic prompts in the Estonian style rather than direct answers. If the child gets stuck for more than two minutes, the system flags it on the coach's screen and the coach walks over to talk it through with them. A short break every twenty-five minutes. After two of those cycles, the morning is done, and the child has done more focused, personally-pitched academic work than they would do in a full week in a conventional NSW classroom.
Twelve hours a week of mastery-tutored English and Maths is more one-to-one academic time than any kid in a normal Australian classroom currently gets. The arrangement just delivers it in the way children actually learn, rather than the way it has been convenient to deliver to thirty kids at once.
“Privacy has to be central, not bolted on later”
The hardware and software have to be flexible and adaptive in the way the system around them keeps changing. Privacy has to be central, not bolted on later. The Department of Education needs to invest heavily in this, and so do the independent and Catholic systems, because the alternative, which is American consumer-grade AI dropped into Australian primary classrooms, harvesting data on six-year-olds and training the next model on their work, is unacceptable in a way that does not need to be argued. The cost of building purpose-built Australian infrastructure is real, and it is also unavoidable.
There are three commitments that are written into the school's charter rather than just its privacy policy. The data stays in NSW. No model is trained on student input. Parents can see, in plain language, every piece of data the system holds on their child, and can request its deletion at any time. An annual independent privacy audit is published openly. These are the conditions for trust, and a school that asks parents to send their children into an AI-rich environment owes them this much by default.
“Half the school day”
The afternoon is the part of the day where AI mostly stays out of the way. The five Cs, anchored in NSW life. Communication, collaboration, creativity, character, connection. Each is taught explicitly, year on year, with measurable progression and external demonstration, in something closer to half the school day than a ninety-minute window per week.
Communication.
Real public speaking is taught from Kindergarten. Five-year-olds tell a story to the rest of the school. Year 3 kids give a three-minute talk at the local cafe to adults who did not have to come. Year 4 starts producing a fortnightly school podcast that parents and the wider community can subscribe to. Year 5 writes letters to the council that get answered, and submissions to government inquiries that get acknowledged. Year 6 stages a TEDx-style event at the end of the year, with talks given to a public audience including local media. Yarning circles run on Mondays, debating on Thursdays from Year 3, and journalism, oral history, and persuasive writing for a real cause are woven through the rest of the week. By the time a child leaves primary at this school they have stood up in front of a room of adults dozens of times.
Collaboration.
Collaboration is taught through real projects with real consequences. Year 5 and 6 run a microbusiness rotation at the Bondi Markets, a different team every other Saturday, with a real budget to acquit and real customers to please. Kids cook lunch for the whole school once a fortnight, by team rotation, from a planned menu and within a real budget. Inter-age peer teaching is built into the week, with Year 6 kids running reading sessions with Kindergarten three mornings each. The school's working market garden is run jointly by Years 3 to 6 in cross-age project teams, and there are conflict resolution roles that Stage 3 kids take turns in. Working together is something kids practise on real tasks, every day, rather than something they get a one-off workshop in once a term.
Creativity.
Creativity is not the specialist subject squeezed into ninety minutes a week. It is closer to the medium in which much of school life is conducted. Every child learns to play one instrument seriously across their primary years, with weekly lessons and daily practice expected. The school produces two full theatrical productions a year, written, designed and directed in part by the kids themselves. Visual arts happens daily, in an actual studio with real materials, taught by a working artist who is around the school often enough that students know what their work looks like. Making is taught, by which I mean woodwork at the safe end, sewing, simple metalwork, ceramics, electronics. Fixing is taught, including bikes, garden tools, broken school furniture, and simple appliances. Cooking is taught not as a one-off but as a rotating responsibility for feeding the school. Photography and film-making run as senior options. Every child by the end of Year 6 has made things, fixed things, performed things, and exhibited things, often enough that the habit is properly formed.
Character.
Character is taught explicitly, rather than left to develop by accident. Grit is taught through specific physical challenges that take weeks to prepare for, with a 2km in Year 1, a 5km in Year 3, a 10km in Year 5, and a half marathon at the end of Year 6. Hard work is taught through extended projects that don't get easier, including a season-long market garden, a six-month research project, and a daily instrument practice log. Kindness is taught through service, with fortnightly visits to the local senior centre over the same children's whole primary journey, peer support roles for younger kids, and structured partnerships with children in special-needs settings. Honesty is taught through dilemmas, with mock trials, ethics circles, and the regular requirement to defend a position you do not personally hold. Perseverance is taught through mastery, where every child commits to one craft, instrument or skill they will work at for a year and demonstrate publicly. These dispositions are taught the way you teach a language, which is to say small and often, with high expectations and consistent feedback, over years.
Connection.
Connection runs in five directions, taught with roughly the same seriousness as literacy.
• To Country. Proper local Indigenous-led learning, not a token acknowledgement. Bidjigal language basics in Kindergarten and Year 1. On-Country lessons with Aboriginal educators, taught where the country itself is, not just in classrooms. Native plants identified by name, used in cooking, traced through their seasons. Dreaming stories taught with cultural protocol. The 8 Ways framework, developed with Aboriginal educators of Western NSW, is the pedagogical anchor for how this work is structured (8 Ways, n.d.).
• To the bush. Annual camps that build in complexity from Year 1 (a single overnight) to Year 6 (a multi-day wilderness immersion). Bushcraft, navigation, fire-building, weather-reading, snake awareness, native plant identification. Not as a once-a-year novelty, but as a recurring rhythm of the school year.
• To the harbour. Sailing instruction from Stage 2 onwards. Kayaking, surf life saving training, marine biology projects with USyd researchers as part of citizen science. By the end of Year 6, every child can swim 1km in open water and handle themselves around a small boat.
• To civic life. Visits to council meetings, courts, parliament, factories, hospitals. Not as excursions, but as regular features of the school year. Every Year 5 student sits through a council meeting at least once. Every Year 6 student visits a working courtroom and writes a piece about what they saw.
• To older Australians. A sustained relationship with one local senior centre, with the same children visiting the same residents fortnightly across multiple years. Oral history projects with elderly Sydneysiders, recorded, edited, and given back as a gift. Older Australians are also invited into the school regularly to teach skills the digital world has thinned out, including gardening, woodwork, mending, story-telling.
• To kids in other suburbs. An exchange programme with a Western Sydney primary and a regional NSW primary, with families hosting in both directions for a week each year. Something more than a school excursion, more like a relationship between schools sustained over years.
These are hours we have been quietly stealing from kids for a hundred years to do timetable management, and they get returned to the children in this design, more or less without apology.
Alpha runs a similar structure in the United States, organised around their own five domains of Teamwork and Leadership, Storytelling and Public Speaking, Entrepreneurship and Financial Literacy, Relationship Building and Socialisation, and Grit and Hard Work. The structural idea is the same as the one used here, but the cultural setting is not. The Australian version foregrounds Country, civic life, the natural environment, and a less individualist take on entrepreneurship than Alpha's TEDx-and-startup-funding model would suggest. By the end of Year 6, our students do not deliver TEDx talks or secure $10,000 of startup funding. They speak at the local council meeting, run a market garden, swim a kilometre in the harbour, sit a yarning circle with an Elder, and acquit a $500 community project budget. The form has been borrowed from Alpha, but the substance is decisively local.
“A child does not graduate without them”
Every Year 6 graduate of this school can demonstrably do the following list of things. These are not aspirational outcomes, they are graduation requirements, signed off in the final term by the child's primary coach, an external practitioner, and a community Elder. A child does not graduate without them.
• Read fluently, write a 2,000-word researched piece, and edit their own work.
• Operate confidently across the full NSW K to 6 Mathematics syllabus, with mastery (not bluff) of every Stage outcome.
• Deliver a ten-minute public talk to an unknown adult audience without notes.
• Lead a yarning circle on a topic of their choosing.
• Swim one kilometre in open water and handle themselves around a small boat.
• Complete a half marathon or equivalent endurance event.
• Cook a meal for thirty people from a planned menu and within a real budget.
• Plan, run, and acquit a community project with a budget of at least $500.
• Play one musical instrument at an intermediate level and perform publicly.
• Make at least one substantial physical object: a piece of furniture, a working device, a sewn or woven item.
• Direct an AI system to do useful work, verify its output, and explain when not to trust it.
• Speak basic Bidjigal language and identify ten native plants on local Country.
• Hold a sustained relationship with at least one older Australian over multiple years.
• Write a letter to a politician that elicits a real reply.
• Articulate, in writing, who they are becoming as a learner.
Fifteen things, concrete, demonstrable, and judged externally. A parent reading the list knows what they are buying into. A principal reading it knows what they would be running. A child reading it knows what is expected of them and, more importantly, why most of it is worth doing in the first place.
“What I came back into the classroom to do”
The teacher's role in this model is not gone, it is elevated. It becomes the thing a chatbot has never done well and most likely will not. Reading a kid's face, holding a high standard when they want to give up, building character, making a class feel like a tribe. We have been wasting our best teachers for years on photocopying and standing-and-delivering, and the new structure puts an end to that. Castlereagh's own language is close enough to the right shape here: redefine teaching as a profession centred on what matters because of AI, including nurturing identity, cultivating purpose, and building the relational and metacognitive capabilities that let learners thrive beyond technology. That is, more or less, what I came back into the classroom to do.
Adults in this model are coaches rather than deliverers. They are paid significantly more than teachers currently are, with smaller groups and higher trust. The teachers I would want at PrimarAI are the ones who are currently leaving the profession because the job has become photocopier-jockey for a curriculum that AI has already disrupted under them. The economics of fewer staff at higher pay actually does work, when the rest of the structure is doing what it should be doing, and it is a better proposition for high-performing teachers than the system they are currently being asked to keep running on goodwill.
The OECD warned in early 2026 about what it called metacognitive laziness, meaning kids skipping the cognitive struggle of learning by hitting prompt-and-paste (OECD, 2026). The 2+4 model is structured against exactly this pattern. The morning forces the struggle, with a tutor patient enough to sit through it, and the afternoon teaches the kid what to do with the brain they are building in the morning. The human in the room, the teacher, is the part of the design that makes the difference between a child who is learning to use AI and a child who is being used by it.
“Six layers, not one grade”
Castlereagh talks about the integrity of assessment, but the phrase is doing too much work. Is the Statement talking about assessment that is more progressive, focused on process and capability rather than product? In the end, isn't it still product that gets assessed, because that is what the system rewards? Isn't this part of the marketisation of education, where assessment is really about status, sorting, and admissions? And why is process so important in the first place, when the process can now be done by AI? Why should kids learn to process things if a machine does it faster?
The honest answer is that none of the current language quite fits. What is needed is a layered assessment system that tells parents, teachers, and the child themselves what they are becoming, what they can do, and where they are heading next. Six layers, each designed to be honest, useful, and difficult to fake.
1. Mastery, not marks
In the morning academic block, every child works toward genuine mastery of the NSW K to 6 syllabus. There are no percentages and there is no 67 per cent. A skill is either mastered or in progress, and the child continues working on it until it is theirs for life. Every Stage outcome is tracked individually, so that parents always know exactly which capabilities have been secured and which are next on the path. Mastery learning is the most evidence-supported approach in education research, and the system uses it for that straightforward reason.
2. Real performance, in front of real people
In the afternoon, children demonstrate what they have learned to actual audiences, including the school community, parents, local elders, cafe customers, council representatives, fellow students. A Year 3 child delivers a three-minute talk at the local cafe. A Year 5 child leads a yarning circle. A Year 6 child runs a community fundraiser from start to finish. The assessment is the doing itself, with no proxy and no shortcut, and children leave this school having stood up and done difficult things in front of real audiences often enough that courage and capability have become part of who they are.
3. The portfolio of a growing person
Every child builds a body of work across their primary years. Their portfolio captures their growth, their range, their best efforts, and the iterations behind them. Portfolios are curated by the child, with the child, and reviewed in regular conferencing conversations. By the time a child leaves Year 6, their portfolio tells the story of who they are becoming as a learner in a way that no number on a conventional report card has ever managed to capture.
4. Knowing yourself as a learner
From Year 2 onwards, students learn to assess their own work, and the assessment of that self-assessment becomes the most important measure of all. Can the child see their own errors? Can they explain why a piece of writing is strong or weak? Can they tell the difference between a confident answer and a correct one? That cognitive capacity is at the centre of what defines a thoughtful adult in the AI era, and the school builds it deliberately from age seven onwards rather than hoping it shows up on its own later.
5. The child as a whole person
Twice a year, every child receives a substantial written narrative from their teacher, which is not a report card but more like a portrait. Honest, careful prose that describes who this child is becoming as a learner, where they are flourishing, and where they need support. Alongside this we report on the dispositions the school exists to cultivate, including curiosity, kindness, honesty, perseverance, courage, and attention. These are the qualities parents most want for their children, and the school is not afraid to name them, observe them, and report on them as part of the formal assessment record.
6. The graduation sign-off
At the end of Year 6, every child is signed off as ready for the next stage by three trusted adults: their primary coach, a community Elder, and an external practitioner in their area of growing expertise. This is the oldest form of assessment in human history, where an experienced person judges a young person ready to move on. We have brought it back because it dignifies the child, dignifies the assessor, and produces a graduation moment that means something concrete. A signed-off graduate of this school stands at the end of primary with a body of work, a list of real things they can do, a clear picture of who they are becoming, and three respected adults who have publicly said they are ready for what comes next.
Why this works
Together, these six layers produce something traditional assessment cannot, which is an accurate, honest, multi-dimensional picture of a real child. The mastery system tells us what they have learned. The performances tell us what they can do in the real world. The portfolio tells us how far they have come. The self-assessment tells us how well they know themselves. The narrative tells us who they are. The graduation sign-off tells us, in the voice of trusted adults, that they are ready.
Our children still sit NAPLAN, and they perform well at it, because mastery-based morning learning develops the underlying skills the test is trying to measure. NAPLAN is one small data point in a much fuller picture, though, and it is not the centrepiece of the school's account of itself. We report enough comparable data that any secondary school can confidently admit one of our graduates, and we are not abandoning measurement, only broadening what gets measured to include the things that actually matter to the child and the people who love them.
“AI for the practice, humans for the meaning”
To show what this looks like in practice, take one syllabus and walk it through. English K to 6 is a good one to start with, because it is where the AI question bites hardest, given that a chatbot can produce a passable persuasive paragraph in five seconds.
Under the current NSW English K to 10 syllabus, students are expected to develop oral language, phonics, reading fluency, comprehension, vocabulary, spelling, handwriting, writing, and creative composition. In the PrimarAI model, the morning AI tutor handles the parts of that list that are genuinely individual and skill-based, including phonics and grapheme practice, decoding, spelling, vocabulary acquisition, sentence-level grammar, reading comprehension at the right Lexile, and oral fluency prompts. These respond beautifully to mastery-based tutoring and adaptive pacing in a way that thirty-kid classrooms have never been able to.
The parts of the syllabus that the AI does not handle live in the afternoon, with a teacher. Writing as a deliberate, iterative, human-led practice. Composition with real purpose for real audiences. Discussion of texts in a circle with classmates and an adult who has read them. Speaking and listening with the actual people in the room. The teacher reads the child's writing line by line, marks it with care, and talks the child through the redraft. None of that is sped up by AI. The AI assists with feedback drafts and idea generation, but it does not replace the human conversation about what the writing is trying to do, and whether it is doing it well.
That pattern, with AI for the practice and humans for the meaning, is the same pattern applied to every other syllabus area in turn. The Blueprint, when published, will work through Maths, Science and Technology, HSIE, PDHPE, and Creative Arts in the same way.
“What a Wednesday looks like”
This section is a description of an ordinary Wednesday at the school, the sort of day a visiting parent or a curious principal might see if they came in and stayed until home time.
The school sits on the corner of two suburban streets. From the outside, it does not look that different from any of the older small schools in the Eastern Suburbs. Inside, the difference is in how the day is shaped, not in the physical fittings, although there is a working market garden out the back and the main hall doubles as a theatre.
Drop-off happens between 8:30 and 8:45, with most kids walking themselves in by Year 2. Bags go on hooks in the tribe rooms, which are multi-age home groups of about a dozen kids spanning Year 1 to Year 6, with a coach who stays with that tribe across their whole primary journey. The tribes gather in the courtyard at 8:45 for a short morning circle, run by a different Stage 3 student each week, where they take the roll, share a thought for the day, and have a moment of quiet before the academic block.
From 9:00 to 11:00, the school is doing its academic morning. Each child is at a workstation in one of three large open rooms set up for different stage groups, headphones on, working with their AI tutor. The tutor knows where each kid sits against every NSW outcome they have done and not done. Mia is in Year 6 today and is working through a tricky bit of MA3-RN-01, which is operations with whole numbers. Her tutor noticed two weeks ago that she had been quietly bluffing her way through long division since Year 4, and she is now backfilling, fast, with the kind of focused attention the system can give her one-to-one that she would never get in a class of twenty-nine. Three workstations along, a Year 1 boy is doing rapid-fire phoneme-to-grapheme practice for EN1-PHOKW-01 because he is behind on his decoding, and the system has him on a slightly different track from his cohort. A coach, fully NESA-accredited and paid more like a professional than the median primary teacher is currently paid, is moving quietly between students, running a small-group conferencing pull-out for kids who have just finished a learning unit, and pulling Mia aside for a five-minute check-in on the piece of writing she has been working on at home.
Nobody is teaching from the front, which is the part most visiting principals notice first. The whole front-of-room setup, which has anchored Western schooling for a hundred and fifty years, is just absent here. The room is also remarkably quiet, in a way that is not the suppressed quiet of a strict classroom but more like the quiet of a small library where everyone is working on something they have chosen, more or less, to care about.
At 11:00 there is recess. Real food, brought from home or made in the school kitchen, and no phones, ever. Kids run around outside on the grass, which is also where most of them eat. The recess break is half an hour, and is the only break in the day where the kids are entirely unstructured.
From 11:30 to 1:00, the day shifts gears completely. This is the public speaking and writing block. On this particular Wednesday, the Stage 2 cohort is rehearsing for tomorrow's session at the Surry Hills senior centre, where they will interview five elderly Sydneysiders about what childhood was like in the 1950s and produce three-minute audio pieces for the local community. The Stage 3 cohort is workshopping persuasive letters to their local council about a missing pedestrian crossing near the school. Kindergarten is in a yarning circle on the rug with their coach and a visiting Bidjigal Elder, who is teaching them a few words of language and a story that goes with them.
Lunch is at 1:00, and on this Wednesday it has been cooked by Year 5, who are on the lunch rotation this fortnight. The menu is sausage rolls and a green salad, with vegetables from the school garden, and the whole school eats outside.
From 1:45 to 3:00 is the project block. This term, the whole-school project is a working market garden at the back of the school, and the work has been spread across cross-age teams of Year 3 to 6 kids. They are building beds, planting natives with a local Bidjigal educator who comes in once a fortnight, running compost data which they will write up as a science report, costing seedlings, and drafting the menu for the parent fundraiser in July. Kindergarten and Year 1 are in the music room with the resident musician, working on a piece they will perform for the parents at the end of the term. Year 2 is in the maker space, halfway through individual woodwork projects which most of them will take home at the end of the term to keep.
The day ends with tribe time from 3:00 to 3:30, where the multi-age home groups come back together for reflection, planning, and the occasional raucous debate about something that happened during the day. The coach who runs the tribe is the same person who was there in Kindergarten and will still be there in Year 6, which is one of the quiet structural things that makes this school feel different from any other.
Pickup is at 3:30, with an extended-care option until 5:30 that is closer to a bushcraft and music studio than to holding-pen babysitting. There is no homework set as a matter of routine. The two-hour academic block in the morning is enough, and the afternoon work, which is essentially the project work most schools would set as homework, is being done with other people, in the world, where it belongs.
A Friday looks different from this Wednesday, and a Tuesday looks different again. Once a fortnight, the whole school goes off-site for the day. The Royal in autumn, the harbour in spring, the courthouse on a parliamentary sitting day, the kitchen of a working restaurant on a quiet midweek. Once a term there is an overnight camp. Once a year, Stage 3 does an extended wilderness immersion in the Blue Mountains, and Stage 2 does a week-long exchange with a Western Sydney primary, with families hosting in both directions. The six hours of the school day are real education, not preparation for an education that happens elsewhere later.
“What makes this school different”
If you only have a minute, here is what makes this school different from every other primary school in NSW.
• Two hours of academic work each morning, instead of six.
• Mastery progression rather than age-grade, where no child moves on until they have it, and no child is held back because the class has not caught up.
• AI used as a Socratic coach rather than an answer engine. Australian-built, NSW-hosted, privacy-first.
• No AI companions or chat-as-friend arrangements, with AI treated as a tool rather than a relationship.
• Multi-age tribes, with the same coach staying with the same children across their whole primary journey.
• Four hours of real-world capability every afternoon, across the five Cs of communication, collaboration, creativity, character, and connection.
• Public performance baked into the year, with every child speaking to real audiences from Kindergarten onwards.
• Real microbusinesses run by the children, with real money and real customers.
• Country-based learning led by Aboriginal educators, not as a token acknowledgement but as part of the regular calendar.
• Annual wilderness immersion that builds from one overnight in Year 1 to a multi-day trip in Year 6.
• Sustained intergenerational relationships with one local senior centre, fortnightly over multiple years.
• No homework, because the afternoon work, done in the world, is the homework.
• Fifteen demonstrable graduation requirements, all of which have to be met before a child graduates.
• Six layers of assessment in place of a single number, producing a real picture of a real child.
• Teachers paid significantly more than they currently are, working with smaller groups, with higher trust placed in their professional judgement.
• Blueprint published openly under a Creative Commons licence, so that any school that wants to adopt the model can.
“Objections, answered”
Will kids fall behind?
They are already behind, by most measures. Three to seven years behind the syllabus on average, in the case of NSW, hidden mostly by grade inflation. Mastery learning, properly implemented, has decades of evidence behind it as a remedy. The Alpha School data is messy and probably overstated in places, but the underlying pedagogy, which is mastery-based progression with adaptive tutoring, is among the most evidence-supported findings in education research. Two focused hours done well looks like it produces better outcomes than six average hours, in pretty much every credible study of the comparison.
What about NESA?
The model is designed to be NESA-registrable from day one as a non-government school. Every minute of the morning maps to specified Stage 1, 2 and 3 outcomes in the K to 10 Mathematics, English, Science and Technology, HSIE, and PDHPE syllabuses. The afternoon four hours covers Creative Arts, PDHPE, and HSIE outcomes more deeply than most schools currently manage. The structure is meeting the syllabus more honestly than the current default does, rather than dodging it.
Won't kids just stare at screens for two hours?
Two hours of focused academic work is less screen time than a NSW Year 4 kid currently gets in an average school day, and far less than what they get on the bus home. The screens at this school are tools rather than feeds, in a closed system with no notifications and no social media. A working session, then off. The four afternoon hours are screen-free unless a child is using AI as a tool for a project, and even then under supervision and as part of a deliberate verification practice.
Does this replace teachers?
No. It pays them more, gives them smaller groups, and frees them from the lesson-prep treadmill. The teachers I would want at PrimarAI are the ones currently leaving the profession because the job has become photocopier-jockey for a curriculum AI has already disrupted underneath them.
What about privacy and Big Tech in classrooms?
Australian-hosted, NSW data-resident, no model training on student data, full transparency to parents on what is collected, and an annual independent privacy audit published openly. The choice is not really between Big Tech in the classroom and no AI. It is between purpose-built AI for kids and kids using ChatGPT on Dad's phone unsupervised, and we are already deep in the second scenario whether we like it or not.
Isn't this just Alpha School with a hat?
No. The school is not selling at $40K a year. It is not screen-streaming kids. It is not American. It is NSW, NESA-aligned, Country-anchored, and public-school adjacent. The thing it shares with Alpha is the maths of the 2+4 structure and a few of the same evidence bases, and almost nothing else.
AI disruption is unknown. How can you plan for it?
You cannot plan for it precisely, and pretending otherwise would be dishonest. Nobody has ten years of experience with this. What you can plan for is the capability set that will be useful across most plausible versions of the next twenty years, which is taste, judgement, agency, embodied skill, relational depth, ethical reasoning, asking good questions, self-knowledge, and synthesis. Build those into a child, and they are about as ready as anyone can be for what is coming.
Does this widen inequality?
If we don't build this, only wealthy children get private AI tutors, AI literacy, and the kind of school day described in this document. If we build it openly, publish the Blueprint under Creative Commons, and design it for NESA registration as a non-government school with a scholarship pathway from day one, the model becomes available to any school that wants to adopt it. The risk of widened inequality comes from not building this, not from building it.
What about kids the model doesn't fit?
This is the most important question in the whole document, and the one Alpha School has handled worst in its own implementation. The answer here is structural rather than aspirational. The AI tutor adapts to twice-exceptional and learning-different kids in ways that no whole-class teacher ever could. Coach-to-kid ratios drop further for those cohorts. Writing instruction is human-led and iterative, not delegated to software. Family engagement is built in from the start. A scholarship pathway from day one prevents the model from selecting only for engaged middle-class families. The school is being designed from the start for the kids the model has to flex around, rather than against them.
“Help me build the next one”
I am one teacher in the eastern suburbs of Sydney with a kid in primary school and a calculator. I can't build this on my own, and there is no point pretending otherwise.
Here is the bigger frame. The Castlereagh Statement is the closest thing Australia has to a national consensus on what to do about AI in education, and it is signed by people I respect, including Danny Liu, Jason Lodge, Leon Furze, Matt Bower, and Simon Buckingham Shum, alongside university leaders, school principals, public servants and Indigenous educators. The Statement is excellent, but it also has a hole in the shape of primary school. The schools-sector chapter is mostly written from a high-school vantage point. The near-horizon actions for primary are sensible and small. The far horizon, where you actually restructure schooling, is left as an exercise for the reader. PrimarAI is me trying to do that exercise in public, showing my working as I go.
What I am going to be doing through 2026 is publishing the full Blueprint in public, fortnight by fortnight. Pedagogy, timetable, syllabus map, staffing model, building, NESA pathway, costs, the lot. Free to read, free to argue with, free to copy. The plan is to align it explicitly with Castlereagh's six principles and three horizons, so that by the time the Australian conversation about K to 6 catches up, there is already a worked example sitting on the table.
If you are a teacher who is tired, a principal who is curious, a parent who is worried, a board chair with a half-empty enrolment, an investor who is interested in what comes next, a Castlereagh signatory who quietly wonders what any of this means for an eight-year-old, or just somebody with a strong opinion and a reason to engage, I want you in the conversation. The work is bigger than any one person can do alone, and the design only really gets interesting when other people start arguing with it.
The model itself is wrong for the era.
Help me build the next one. With Country, not on top of it.
With teachers, not around them.
With children, who, believe me, already know.
Dane Smale, Sydney, May 2026
Licence
© 2026 Dane Smale. Licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).
You are free to share and adapt this work for any purpose, including commercial, as long as you credit the author and link to the licence.
https://creativecommons.org/licenses/by/4.0/
Cite as: Smale, D. (2026). A practical plan for an AI-native Australian primary school. PrimarAI. https://primarai.com.au/