In April, McKinsey published an article called "Where AI will create value — and where it won't." The argument was sharp: AI is not a productivity revolution, it is a competitive reset. The companies that win the next decade will not be the ones that adopt AI fastest. They will be the ones that understand earliest where value is moving — and position themselves at the new control points before the equilibrium solidifies.
McKinsey wrote that piece for business. This one is for education.
The same three-wave logic applies to American public schools, with surprising precision and a deeply uncomfortable conclusion: the institutional layer of K-12 — districts, county offices, state agencies, the trade associations that serve them, the governance-technology vendors who have quietly built unusually attractive multi-state businesses on the friction of policy compliance — is more exposed to structural reorganization than almost any sector outside finance and retail. The window to position is open now. It will not stay open.
The Three Waves, in K-12
McKinsey distinguishes three overlapping ways AI creates economic value. The first is productivity — automating tasks, accelerating existing work. The second is differentiation — new products, services, and business models that were previously infeasible. The third is transaction-cost reduction — the radical compression of search, comparison, switching, and coordination costs through agentic AI, which Coase's transaction-cost economics predicts will reorganize entire industries from vertically integrated value chains into modular, agent-mediated ecosystems.
The McKinsey verdict is that productivity gains rarely produce durable advantage because competitors match them, and the benefits flow to customers through lower prices rather than to the firms that adopted the technology. Differentiation produces moats but only when protected by proprietary data, network effects, or business-model innovation. Transaction-cost compression — the third wave — is where market structures themselves change, where intermediaries are bypassed, and where value migrates to whoever occupies the new control points.
The K-12 mapping is unusually clean.
Wave 1 in K-12 is teachers using AI tools — MagicSchool, Brisk, Curipod, Khanmigo, Diffit — to draft lesson plans, build rubrics, generate IEPs, and triage parent emails. Districts adopt the tools; teachers get hours back; the public is faintly impressed. Almost every district in America is here. Almost none has moved past it. McKinsey's warning lands hardest at this wave: a district that adopts MagicSchool is not materially better positioned than the district next door that adopts Brisk. The teacher gets time back. The vendor captures subscription revenue. The district has built no durable advantage.
Sal Khan said as much in April. Asked about Khanmigo's classroom impact, the founder of Khan Academy acknowledged that for most students the AI tutor has been a non-event — that an AI tutor in the back of the classroom is like a human tutor in the back of the classroom. Some students will seek out help. Most will not. AI tools by themselves do not generate the motivation or metacognitive awareness that drives learning gains. They are infrastructure, not magic. Khan Academy has begun rebuilding its district platform around motivation and human systems, which is the right move and which also represents a quiet retreat from the AI-as-headline-feature posture of two years ago. It is the most honest single statement in K-12 AI right now.
"The K-12 sector's current obsession with productivity AI is the educational equivalent of replacing the steam engine with an electric motor while leaving the line-shaft layout untouched."
Where Wave 2 is Actually Happening
The most aggressive Wave 2 experiments in American education are not happening inside traditional public schools. They are happening at the boundaries — in private and charter networks, in microschools, and in the rapidly expanding world of education savings accounts.
Alpha School, the private K-12 network founded in Austin in 2014, has built a model called 2 Hour Learning. Students do focused academic work each morning on an AI-driven instructional platform called Timeback; the rest of the day is devoted to life skills, entrepreneurship, public speaking, athletics, and project work led by adults the school calls guides rather than teachers. Alpha now operates campuses in Austin, Plano, Chantilly, Miami, Palo Alto, and other cities. Tuition runs from approximately ten thousand dollars in Brownsville to seventy-five thousand at the highest-priced campuses. Alpha claims its students score in the top one to two percent on third-party MAP Growth assessments and progress at roughly 2.3 times the national norm. Alpha is also pursuing public-sector entry through a virtual charter authorization in Arizona and a physical charter application in Texas.
It is essential to be honest. Alpha's growth claims have not been independently verified by peer-reviewed research. The model is financially unavailable to most American families at its current price points. None of that, however, refutes the strategic point: Alpha is the most visible attempt yet to redesign the cost structure and time architecture of schooling around AI as the primary instructional engine, rather than as a supplement to a traditional teacher-led classroom. It is the most consequential Wave 2 experiment in K-12, and it is reaching the public sector through charter and ESA authorization, not through the traditional district pipeline.
The education savings account expansion is the policy mechanism through which Wave 2 reaches scale. Florida operates the largest ESA ecosystem with over five hundred thousand students. Arizona's Empowerment Scholarship Account program reported 102,359 students enrolled for 2025-26. Iowa moved to universal eligibility for 2025-26. Texas's newly enacted ESA program received over 100,000 applications in less than two weeks for its inaugural fall 2026 launch, with an initial appropriation of one billion dollars. Tyton Partners reports that more than 92,000 students participated in ESA programs nationwide in 2023-24, with roughly $680 million flowing out of the public K-12 system through these accounts. Two and a half years later, those numbers are an order of magnitude larger.
The strategic significance of ESAs is that they convert the educational consumer relationship from "enroll in one school" to "assemble a bundle of services." Once bundling is legal, AI-mediated providers can compete on individual line items rather than on whole-school enrollment. This is the Wave 2 product-frontier expansion that McKinsey describes — except in K-12 it is being accelerated by policy design, not by technology alone.
The Coasean Compression Comes for Public Education
Wave 3 — transaction-cost reduction — is where the McKinsey framework gets uncomfortable for K-12. Public education is, by structural design, drenched in transaction costs. Almost every layer of intermediation in the sector exists because some form of friction exists at the layer below it.
Curriculum adoption cycles run six to seven years, governed by state board adoption frameworks, district committees, vendor pilots, and board approval. Procurement of edtech and instructional materials is governed by state public-contract code; an RFP cycle can extend to twelve to eighteen months. Compliance reporting consumes enormous administrative capacity — Local Control and Accountability Plans, School Accountability Report Cards, Title I and IDEA documentation, dozens of categorical-funding accountability artifacts. Family-school information asymmetry is a foundational design feature of district communication. Switching costs for families — residency requirements, transcript portability, IEP transitions — are deliberately high. The intermediary stack — county offices of education, school boards associations, regional educational service agencies, policy services, student-information-system vendors, assessment vendors — exists in part to manage all this friction. Each layer makes a margin proportional to the friction it manages.
If AI agents materially reduce these costs at near-zero marginal cost — what researchers have begun calling the Coasean singularity — the entire sectoral architecture is exposed to reorganization. Five concrete shifts are already legible:
- Curriculum procurement compresses, because an AI agent that continuously evaluates every available curriculum against state standards, district demographics, and outcome data does not need a multi-year adoption cycle.
- Compliance becomes continuous, because regulatory artifacts can be maintained as living documents rather than assembled annually.
- Family-school discovery reverses, because in an agent-mediated world an AI representing a specific child evaluates every accessible educational option continuously, making brand recall and marketing investment matter less than algorithmic ranking and structured data quality.
- County and regional intermediaries face an existential question, because the information-arbitrage portion of their portfolios is the portion most directly exposed.
- Multi-state governance-technology vendors who have quietly built unusually attractive businesses on the friction of policy compliance, meeting orchestration, and document control find their friction-management moat narrowing — exactly the friction that AI agents are progressively compressing.
The Coasean argument is not a prediction. It is a description of force gradients. Where transaction costs fall, value migrates. Where value migrates, control points relocate. Whoever occupies the new control points first captures the durable advantage.
Five States Are Already Moving
The calendar is shorter than it appears. At least five states currently have proposed or enacted legislation governing AI in K-12 education at significant scope:
- California has the SB 1288 model AI policy framework, requiring the State Department of Education to produce a model AI policy for districts in 2026, with twenty-five operative obligations on the implementing layer.
- Arizona has AI policy and curriculum-integrity provisions tied to its expanding Empowerment Scholarship Account program.
- Florida has issued AI policy guidance through the Department of Education with pending legislative action expanding the framework.
- Idaho's S.B. 1227 directs a comprehensive generative AI framework for K-12.
- Texas has AI-related provisions tied to its new Education Savings Account program and complementary legislation moving through the 2026 session.
More than thirty states have released AI guidance for schools. Eighteen state legislatures are actively debating major K-12 AI bills. At least fifteen states have enacted or proposed mandates that districts adopt formal AI policies on specified timelines. Every one of these legislative actions creates a new compliance surface that did not exist twelve months ago. Every district subject to any of these statutes will, by 2027, need to produce, maintain, and demonstrate compliance with AI-related policy and operational requirements.
The Equity Question McKinsey Did Not Address
There is one place where the McKinsey framework, applied to K-12, breaks down — and it is the most important place. The McKinsey article contains essentially no discussion of distributional consequences. In K-12 that omission is not optional, because the families who benefit most from agent-mediated discovery are precisely the families with the resources to deploy sophisticated agents on their behalf — families with money, English fluency, time, and policy literacy. The families who benefit least are exactly those whom the public education system is most committed to serve.
An ESA-funded, agent-mediated education ecosystem, designed without intervention, will expand opportunity for families who already have it and concentrate the disadvantages of those who do not. This is not a theoretical concern. The architecture being built right now will determine which children are visible to the agent layer and which are invisible. A child whose family has the literacy to deploy an AI agent that continuously evaluates educational options will get options. A child whose family does not will get whatever the default is. In an agent-mediated world, the default is increasingly assigned by the absence of an advocate.
"The most strategically and morally defensible response is to build the infrastructure that ensures every child — not just the well-resourced child — is visible to the agent layer."
That means measurement of subgroup outcomes that the existing system cannot see. Verification of whether intervention spending actually reaches the students whose specific deficits the district has flagged. Public data feeds that AI agents can read on behalf of every child rather than only on behalf of well-resourced families. These are not abstract policy goals. They are concrete operational decisions being made now, in the design of state dashboards, in the procurement requirements being written into AI-related legislation, and in the technology choices districts and county offices are making for the next academic year.
The Window
The most important sentence in the McKinsey article, applied to K-12, is the one near the end: history is littered with examples of companies that mistook efficiency for advantage. They optimized while others reinvented. They cut costs while others captured dominant market share. When the dust settled, the winners were not those who adopted the technology fastest but those who understood where value was moving earliest, and positioned themselves to capture it.
Translated to K-12, the equivalent observation is that the districts, county offices, state associations, governance-technology vendors, and edtech firms that adopt AI tools fastest will not be the winners of the next decade. The winners will be those who understand that the underlying market structure of public education — the relationship between families and schools, the role of intermediaries, the architecture of measurement and accountability, the procurement and policy layers — is exposed to a kind of structural reorganization the sector has not faced in a hundred years.
AI will not create value evenly in public education. In many cases, it will simply redistribute it. The window to act is narrower than it may seem. In an AI-driven economy, advantage compounds early — and value capture locks in fast.
The decisions that will determine who occupies the new control points are being made now. Five states are already legislating against the substrate that does not yet exist. The technology to build the new measurement, verification, and accountability infrastructure exists today. The institutional question is whether the sector chooses to build it deliberately and equitably — or whether it allows the agent layer to be built, by default, around the families who already have advocates.
There is a third option, which is to wait. That is the option McKinsey's history of general-purpose technologies suggests will be punished most severely. The window is open now. It will not be open indefinitely.
———