Children who start kindergarten next fall will graduate in 2040. This is a plan for what school must give them.
School must build a mind capable of standing without AI — and a person capable of accomplishing far more with it.
Children who start kindergarten next fall will graduate in 2040.
School must build a mind capable of standing without AI — and a person capable of accomplishing far more with it.
This is a draft, not a doctrine. Build what students cannot safely outsource. Teach them to use what can be outsourced. Protect what should never be outsourced.
Kindergarten in fall 2027. Diploma in June 2040. A composite child, not a real one — she'll walk through this talk with us.
We can't know what job Maya will have. But we can already see that the world she'll learn, work, and live in is changing.
Every decision in this room is about children who already exist — with names, bedtimes, and thirteen school years ahead of them.
Serious people disagree about the pace. The honest position is a range.
Slower — many economists' base case. AI stays a powerful copilot. Jobs, institutions, and robotics change gradually.
Median — my best guess. AI agents absorb much routine desk work. Roughly half of adults keep conventional jobs; perhaps one in four run their own agent-powered work.
AI 2040 — a stress test, not a forecast. Machine thinking, then machine labor, becomes nearly free. Radical economic and social change.
The honest position is a range.
A year-by-year scenario of the road to superintelligence — extraordinary detail on compute, treaties, safety, agents, robotics, distribution. It barely discusses what happens to school.
Its authors call it primarily a recommendation, not a prediction — a deliberately extreme stress test.
Someone still has to write the education chapter. Tonight, we draft it — for all three futures at once.
One reasonable 2040 labor market — my guess, not data.
About half of adults may not be in a conventional job — 'unemployed' by traditional measures. About one in four may be creators, founders, and operators of one-person, agent-powered businesses. The rest retired or out of the labor force.
Many adults will work less like employees and more like small companies of one — directing teams of AI agents instead of reporting to a manager.
When expertise is abundant, the scarce skill isn't knowing the answer. It's choosing the question, setting direction, and judging the result.
We're not just changing how students learn. We're changing what school is for — from producing employees to developing people who can direct intelligent systems.
The economies differ dramatically. The educational bets do not.
If AI 2040 turns out to be twenty years early, every change proposed here still makes school better now.
AI can already produce the essays, problem sets, and code we grade. The work no longer proves the learning.
Education 1.0 taught facts. Education 2.0 taught skills. Education 3.0 must teach judgment.
These are complementary, not competing school models.
Reading, writing, math, science, history, arts, civics — not to memorize everything, but to build an internal model of how the world works.
Enough structure to question, connect, and doubt — so a student can spot errors, judge evidence, think without help, ask better questions, resist manipulation, and take part in civic life.
You can't spot a confident hallucination in a subject you never learned. Reading and writing are technologies of thought — not delivery formats to retire.
Foundations — math, science, history, arts. Building an internal model of the world with enough structure to question, connect, and doubt.
Studios — interdisciplinary projects with AI agents on the team, building real software, businesses, research, and art for real audiences.
Knowledge feeds the projects. Projects give the knowledge purpose.
The project is the vehicle. The real curriculum is judgment, taste, ethics, collaboration, communication, reflection.
A mind that can stand without AI — and a person who can accomplish far more with it.
A graduate can frame a meaningful problem, direct people and machines toward it, verify the result — and own the consequences.
AI is present all day — and the center of none of it.
School becomes more human because the infrastructure becomes intelligent — teachers' hours go to dialogue, critique, mentorship, and culture.
Each question gets its own evidence — kept separate on the transcript.
The future must not become wealthy children receiving exceptional AI, teachers, and experiences — while disadvantaged children receive a surveillance chatbot.
Human-rich. Privacy-protecting. Equitable — by design, not by afterthought.
Begin where the leverage is: let AI absorb selected practice, variation, and first-pass feedback — and give the reclaimed hours to explanation, dialogue, critique, mentorship, and culture.
The Class of 2040 must graduate able to think alone, build with others, direct machines — and own the consequences.
AI may arrive on AI 2040's timetable — or decades later. But the Class of 2040 gets only one childhood.
We don't need to know the economy of 2040 to know what the Class of 2040 will need.
AI may arrive on AI 2040's timetable — or decades later. But the Class of 2040 gets only one childhood. We don't need to know the economy of 2040 to know what the Class of 2040 will need.