AI in Education · A project line
The question is not whether AI belongs in schools, but when.
What is coming, and what the science of learning tells us about how to meet it.
The asymmetry of risk
The cost of getting the timing wrong is not symmetric.
Scroll the chart sideways to see both endpoints →
Illustrative. The shape of the argument, not a measurement; no dataset yet traces these curves.
Introducing AI to a learner later than they could have used it carries a real cost, and a recoverable one. Some efficiency is lost; a tool is adopted a little after it might have been.
Introducing it earlier than a developing mind can absorb carries a different kind of cost. It may fall on the formation of the mind itself, during the years that mind is still being built. That cost may not be recoverable.
A school that has waited has not failed; it has a sound place to begin. A school where AI is already in use, perhaps further and earlier than was ever decided, has something to bring structure to.
Being first is the headline. Being right is the foundation.
What the work rests on
Readiness is a sequence, not an age.
Scroll the chart sideways →
01
Development is staged, so the teaching of AI should be
A knowledge base, the automatisation of basic academic skills, and the capacity to hold information in mind all consolidate early. Judgment matures late; the systems that govern it develop into the early twenties. A capacity still being built cannot safely be handed to a tool, because handing it over removes the effort that builds it.
02
Responsible reasoning is possible before the evidence settles
We reason from the developmental sequence and from earlier technologies that changed the cognitive work a learner does. Calculators and search engines are instructive and imperfect analogies; each offloads a bounded task, where a general AI can produce the whole artefact. We weigh those cases only where they genuinely resemble AI.
What the science establishes is the sequence: understand before operate, depth before independence, foundations before offloading. What a practitioner decides is where those boundaries fall in particular school years. The first is grounded and we defend it. The second is a reasoned, revisable translation, and we name it as one.
Research Report · RR-AI-01
Adoption is no longer the question.
A generation now uses AI daily that nobody taught them to use.
~700M
ChatGPT weekly users by mid-2025, near one in ten people on earth.
54.6%
of US working-age adults had used generative AI by August 2025.
77%
of all use is three everyday tasks: practical guidance, writing, and seeking information.
Position Paper · PP-001 · AI by Design
The Formed Mind
Our developmental framework for AI in schools. AI should extend a developing mind rather than replace the effort that forms it, and support a young person’s relationships rather than stand in for them. Free for school leaders.
Why we work on this
A childhood is the unit of measurement, and we are inside one.
There is a generation of children in school now. The evidence that would tell us with any certainty how AI use shapes a developing mind will not arrive for fifteen or twenty years. Most writing on AI in education either waits for proof it cannot yet have, or proceeds as though the proof were already in. We do neither, because schools are making decisions they cannot defer and deserve reasoning that is honest about its own limits.
Forecast Companion · FC-AI-01
What Follows
The companion to The State of AI Use: four extrapolations the trend lines will carry, one open question they will not, and what each asks of a school — and of the companies selling to them.
Articles in this series
The EdJournal Advisory
When is a child ready for AI?
If that question is live in your school, the Advisory helps leaders translate the developmental approach into a framework their own school can run.