Every child in Wales carries a key (and deserves to find it).

Gareth Milton Griffiths, founder of Athro.ai and Fflam

I want to tell you about a moment that happens in Welsh classrooms with increasing regularity. A teacher — good teacher, prepared teacher, teacher who cares — asks a student how they’ve been revising. The answer comes back without hesitation: “I just ChatGPT it.”

Three years ago, that answer would have raised eyebrows. Now it barely registers. It has become the background noise of Welsh secondary education — a new kind of non-answer to an old and important question.

Standing at the front of Welsh classrooms as an educational presenter, I watched it happen over and over and felt something I can only describe as professional grief.

Technology is neutral, so it was nothing to do with students using tech and it’s not even because they were cutting corners — students have always cut corners and always will. But because of what was being skipped and because of what was being lost. That’s where the professional grief lived.

There is a ladder that runs from the first spark of curiosity about a topic all the way to the moment a student puts pen to paper in an exam room, or produces something that is genuinely, unmistakeably their own. Every rung on that learning ladder matters. The moment of confusion that precedes understanding. The process of finding your own words for something difficult. The satisfaction — and it is a real satisfaction, even for reluctant learners — of realising you actually know something now that you didn’t know before.

That ladder was being skipped altogether.

Not climbed faster. Not climbed more efficiently. Skipped. Students were arriving at the output — the essay, the answer, the revision note — without having made the journey that gives the output its meaning. And the AI tools enabling this weren’t malicious. They were simply indifferent. They were built for someone else, somewhere else, and they had no stake in whether a fifteen-year-old in Carmarthen actually understood what they were submitting.

Before I continue, I want to say something about keys.

Every learner carries one. It is shaped by the way they think, the pace at which ideas settle, the language that feels most like home, the kind of question that makes them lean forward rather than sit back. It looks different for every child. It is different. And somewhere inside each of them is a cathedral of knowledge — mathematics, literature, science, history, the whole interconnected architecture of human understanding — that the right key can open.

What I watched happening was learners being handed keys cut for someone else’s lock. The door stays shut. And after a while, they stop trying.

That is not a technology problem. It is an equity problem. And it will not be solved by banning the tools that expose it.

I spent ten years in Welsh classrooms before I understood that the answer wasn’t to resist the wave — it was to ask why it was being ridden so badly, and whether anyone had thought carefully about building something better. I spent ten years before that as a web designer and digital developer, watching technology reshape every industry it encountered.

Those two decades converged into a single question:

What if the tools being used in Welsh classrooms were actually built for Welsh learners — rooted in the Curriculum for Wales, bilingual by design, shaped around Welsh values and Welsh identity? Would the ladder still be getting skipped?

I decided to find out.

Socratic

The approach that emerged is Socratic. Not because that word sounds impressive, but because it describes something precise and important: a tool that does not give answers, but helps learners find them. That asks what a student already understands before it adds anything. That guides rather than delivers. That keeps the learner in the act of thinking, because that act is the learning itself.

But Socratic dialogue is only one instrument in the room.

Imagine a system — and I want you to hold it as an idea before anything else, because what follows is not a list of features but a description of what it looks like when technology is genuinely built around the learner rather than around the sale.

A student arrives with homework and no idea where to begin. Before anything else, the system asks what they already know. The answer — however partial, however wrong — is the starting point. From there, the learning is built outward from what the student actually has, not from what the curriculum assumes they should have.

When something needs to be seen rather than read, a simulation opens — not searched for, not linked to, simply there, because the system understood what was needed before the student thought to ask.

When a student is ready to go deeper, a full course can be generated in seconds — lessons, slides, immersive text, quizzes — drawn from peer-reviewed research and curriculum documentation, structured around exactly what they are studying, available in whatever format suits how they think.

Notes are generated after each session: not answers to be lifted, but a quiet distillation of what was explored and what still needs to come back.

Flashcards are rated by the student’s own confidence. Spaced repetition surfaces the difficult ones first. A study plan is built around the gaps in a student’s calendar and the gaps in their understanding — and when they ask what to work on next, the system tells them, because it has been paying attention the whole time: tracking effort, measuring confidence, noticing the topics that keep getting avoided.

When the exam approaches, past papers are available in two modes. In one, each question is worked through with feedback at every step — the thinking stays with the student, the scaffold stays in reach. In the other, it disappears entirely: timed, unassisted, exactly as it will feel on the day.

A map

What follows is not just a mark. It is a map — what went well, what needs work, what to study next, and precisely what it would take to reach the next grade.

The key changes shape to fit the lock. That is the only design principle that matters. And it is the one that has been missing from every tool built for a different country’s classroom and handed, with the best of intentions, to a child in Wales.

I am not describing this to sell it. I am describing it because every part of it was designed to answer the same question: what does a child actually need to climb the ladder themselves? The answer, it turns out, is different for every child. That is the whole point.

And for teachers — who carry a weight of planning, marking, and administration that has bent many of the best people in Welsh education toward leaving the profession — the same intelligence works differently but with the same purpose.

A lesson plan, built around the Curriculum for Wales, in minutes rather than hours. Homework constructed to the exact progression step of the class. Marking frameworks generated and returned.

Not replacing the teacher’s judgement — that judgement is irreplaceable — but restoring the hours that judgement deserves to occupy. The hours currently spent on paperwork that could be spent on the child sitting in front of them.

athro.ai

This exists. It is bilingual. It is live. It was built in Wales, for Wales, for every school in both languages, because that is what Wales actually looks like. It is called athro.ai and I built it from the ground up.

Two pieces published on Nation Cymru recently described the problem with clarity and care.

A consultancy communications lead wrote about Welsh teachers spending their Sundays adapting tools built for England’s National Curriculum, or America’s Common Core, because nothing purpose-built exists. A former minister echoed it: Welsh schools may already be spending a million pounds a year on AI software that simply displaces one task with another without improving any outcome that actually matters.

Both were right about the gap. And one passage in particular deserves to be quoted directly, because it names something real:

“What is missing is the infrastructure that turns a genuinely good working example into something the whole system can rely on. A working example needs an identity. Positive feedback from teachers needs to become evidence an official can assess. A proof of concept needs governance, sustainability, and a plan for what happens as the curriculum evolves. Those things do not build themselves.”

This is exactly right. And I want to respond to it not with a rebuttal, but with a recognition — and perhaps an offer.

That work is precisely what Wales needs to begin. Turning the feedback of Welsh teachers into evidence that officials and ministers can stand behind. Working through governance — who owns it, who maintains it, how it evolves as the curriculum evolves. Mapping the sustainability route through funding mechanisms that already exist within Welsh Government.

Setting out a credible plan a minister can champion. Not just a current state a minister can tour.

That is the conversation Wales needs to have right now, in this week, as a new government forms and decisions about digital education infrastructure are being shaped by what is visible and what is not.

In times when there might have been every reason to doubt whether something like this was worth attempting — when building bilingual, ethically-designed AI for Welsh learners felt like a task too large for one person — the question was never whether to stop. It was always: how do I make sure this reaches the people who need it?

Shaped

There are children in classrooms across Wales right now with cathedrals inside them. The keys exist to open the doors to those cathedrals — Socratic, adaptive, bilingual, built for the Curriculum for Wales, shaped around the learner in front of them rather than the learner someone else imagined. They should not be hidden. Not from a single child in a single classroom in this country.

What is missing is not the platform.

What is missing is the bridge between a thing that works and a thing the whole system can depend on. That bridge is a conversation. And this is the beginning of it.

No child left behind.

Gareth Milton Griffiths is the founder of Athro.ai, a bilingual AI tutoring platform built for the Curriculum for Wales, and Fflam, a study skills workshop business serving Welsh secondary schools. He has spent ten years as an educational presenter in Welsh schools and twenty years as a digital developer. He is from Aberteifi and lives in Llantrisant.