In April, I wrote in these pages that AI is not a productivity revolution but a competitive reset, and that the institutional layer of American K-12 education — 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.
I argued, applying the McKinsey three-wave framework to schools, that the window to act is open now and will not stay open.
The three waves were:
- Productivity — what teachers using MagicSchool or Brisk get: hours back, no durable advantage, the benefits flow through to customers.
- Differentiation — what Alpha School is attempting at the boundary of the public system, redesigning the cost structure and time architecture of schooling around AI as the primary instructional engine rather than as a supplement.
- Transaction-cost compression — the Coasean wave — where the sector's friction-managing intermediaries face the existential question, because the cost of managing information asymmetry is the business most directly exposed to agentic AI.
That was the argument. What I did not have in April, and what I have now, is the canonical analog from another regulated, multi-stakeholder, life-shaping sector that has already begun to make the transition the McKinsey framework describes. The analog is medicine, and the canonical voice is Stéphane Bancel, the CEO of Moderna, in conversation with McKinsey's North America chair in late May.
Biology as Information
Biology is information. Not biology as biochemistry, not biology as molecular interaction — biology as the flow of information.
"When you die, you still have information in your DNA," Bancel says. "We can retrace DNA in bones of people who have been dead a long time. But to stop the flow of information is basically death in biology."
Cancer, in this frame, is instability of human DNA. Some instabilities are caused by genetic mutation, some by viruses, some by external factors like sun exposure that drives melanoma. Treatment becomes the act of giving the immune system information about which specific instability to correct — different information for different patients with the same diagnosis, made through the same manufacturing process.
This is a profound reframe of what drug development is. "The pharmaceutical drugs that have affected billions of people in the last 100 to 150 years came from what was essentially an analog world," Bancel says. "You threw spaghetti on the wall, saw what stuck, and what worked, and then took those solutions to patients."
That sentence describes the entire industrial era of medicine. It also, with minimal translation, describes the entire industrial era of public schooling.
Education Is Also Information
A student's learning trajectory is information. The achievement gap is the measured distance between intended information transfer and actual information transfer.
The English learner whose oral English is two grade levels ahead of her written English is an information-flow mismatch — and the absence of any instrument in California to detect that mismatch is the educational equivalent of having no biopsy, no biomarker, no early-stage screening protocol. The child grows. The mismatch compounds. The intervention, when it finally happens in eighth or ninth grade, costs far more than the early intervention would have.
"Biology is an information-based science. It's only information." — Stéphane Bancel
Bancel says this twice in the McKinsey interview, in two separate passages. It is the load-bearing claim of the entire mRNA program.
Once you accept that biology is information, the rest of Moderna's architecture follows: a single manufacturing process that produces different molecules from different informational inputs; AI as the natural mode of iterating on the information substrate; a feedback loop where robots run experiments and the data goes back into the system to design the next experiments; and an organizational design built around the proposition that human-in-the-loop must be reinvented because the work itself has changed.
Now read that paragraph and substitute education for medicine, classroom for clinic, district for hospital. The argument holds without a single forced parallel.
Education is information. The work of teaching must be reinvented because the substrate has changed. Measurement must function as the biopsy, the biomarker, the early-stage screening protocol. Intervention must be different for different children with the same nominal diagnosis. The infrastructure that makes this possible must be approved as a process, not as a product, because every district will need a different specific output from the same underlying instrument.
The End of Average, in Medicine and in School
Eric Kutcher, McKinsey's interviewer, names the shift directly: this is the end of average.
Bancel agrees. "If you and I are diagnosed by the same doctor in the same hospital on the same day with skin cancer, we will now make a drug for you and a different drug for me. The drugs will be chemically different."
That is the operational meaning of personalized medicine. The same diagnosis produces different molecules because the underlying instability — the underlying information — is different in each patient's DNA.
The educational parallel is exact. Two third-graders in the same classroom, given the same nominal diagnosis (struggling reader, English learner, intervention candidate), require different specific instruction because the underlying information-flow mismatch is different in each child. One has strong oral comprehension but weak written production. One has the opposite. One reads phonetically but cannot infer meaning. One infers meaning but cannot decode unfamiliar words.
The industrial era of schooling gives them the same intervention — "more reading instruction" — because the system could not see the differences. It still cannot, in most California districts, because the measurement infrastructure that would make the differences visible does not exist.
Bancel's argument about cost compounds the point. "When you get immunotherapy today, five years out from the start of treatment, only 50 percent of patients have responded well. If you think of it from a cost standpoint, half of the cost of a drug went into the garbage."
Apply this to the eighty billion dollars California spends annually on K-12 education. We do not know what fraction is reaching the specific deficit it was intended to address. Districts cannot tell us because they cannot measure at that resolution. The state cannot tell us because the accountability architecture aggregates to the school level, not to the child level. We are spending the money. We do not know which half went into the garbage.
The Process, Not the Product
This is where Bancel's argument becomes most directly useful for education policy.
When every patient gets a different drug, you cannot get the drug approved in the traditional sense, because the drug for you and the drug for me would be different chemical inputs. So Moderna gets the manufacturing process approved. "I have to prove to a regulator that every time I get the same input, the same information about somebody's DNA, they get the same output, the same medicine."
California's educational governance has been trying for thirty years to approve the product.
It has tried Standardized Testing and Reporting, the Academic Performance Index, the California Standards Tests, the Smarter Balanced Assessment Consortium, the Local Control Funding Formula, the California School Dashboard, the Local Control and Accountability Plan. Each one was an attempt to standardize a product — a single accountability measure, a single dashboard, a single intervention category — that every district would deliver to every child. Each one failed in the same way: the product worked for the average child in the average district and missed the specific child in the specific district who needed the specific thing.
The achievement gap did not close. After thirty years it has not closed.
AB 2225, which is moving through the California Legislature this spring, is the first piece of legislation we have seen that points toward the right answer without yet naming it. The bill instructs the State Board of Education to convene a Working Group that will produce a Plan by December 2027.
If the Working Group produces another product — another standardized accountability frame — the next thirty years will look like the last thirty. If the Working Group produces a process — a measurement infrastructure that every district feeds into and that every district receives differentiated outputs from — then California will have done in education what Moderna has done in medicine. The same input produces the same output. Different children produce different interventions. The state approves the process. The districts run the patients through it.
Here Is What Must Change
What mRNA has accomplished in medicine is what must take place in American education. Not the molecules — the architecture. Not the specific therapies — the substrate. The operational changes that follow from taking that parallel seriously are concrete, and the technology to make them exists today.
Measurement infrastructure must be treated as the substrate, not as an afterthought. California spends roughly one-half of one percent of its K-12 budget on assessment. The figure should be three to five percent. Without a measurement substrate, the rest of the architecture cannot function — districts cannot tell which intervention worked, the state cannot tell which districts need help, families cannot tell which schools are actually serving children like theirs. Moderna does not run a clinical trial without a manufacturing process. California should not run an education system without a measurement substrate.
Prevention must replace remediation as the dominant economic logic. Bancel: "detecting cancer in stage one of the disease, when it's very early, increases the odds and thus reduces the cost of treatment." The educational analog is the third-grader whose oral-written delta flags before the literacy gap compounds, the fourth-grader whose math reasoning outpaces her arithmetic fluency before she gives up on math, the seventh-grader whose social-emotional indicators move before the disengagement becomes truancy. Each of these is detectable now, by instruments that exist. The state has chosen not to deploy them at scale. That choice is the educational equivalent of declining to screen for cancer.
Every child must be visible to the agent layer that is about to be built. In an agent-mediated world, 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. The remedy is not to slow the agent layer. The remedy is to ensure every child is in the data — measured, visible, reachable by an agent acting on her behalf — whether or not her family has the resources to advocate for her. This is what public infrastructure means in an agent-mediated economy.
Intervention must be approved as a process, not as a product. The model policies California is producing in 2026 — including the SB 1288 model AI policy with its twenty-five operative obligations — are still trying to approve the product. They specify what the LEA shall do, how it shall be governed, what consent shall be obtained. They do not specify the measurement instruments by which we will know whether any of it is working. AB 2225 is the vehicle through which the process — measurement, verification, accountability — can be approved.
The work itself must be reinvented. Bancel's most striking observation about Moderna's organizational design is that the head of HR is also the head of digital, because in an AI-and-robotics world the central question is not how to govern AI tools but how to reinvent the work humans do. The educational parallel is the teacher's day. The industrial-era classroom assumed one adult orchestrating one mode of instruction for thirty children across six hours. The post-industrial classroom requires the teacher to orchestrate multiple AI-augmented modes simultaneously — adaptive instruction, intervention triage, family communication, social-emotional monitoring — with the AI doing the parts machines do well and the teacher doing the parts only a human does. Every district has Section 22 of the new model policy in its policy binder, requiring educators to adapt lessons using AI-based tools. None has rewritten the workday in which that adaptation becomes possible. That is the reinvention the next decade requires.
The Decision Is Now
The April article ended with a warning that the window is narrower than it appears. The mRNA analog sharpens that warning into a specific obligation.
Medicine has shown what the next era of a regulated, multi-stakeholder, life-shaping sector looks like when its central insight — that the substrate is information — is taken seriously and the infrastructure is built deliberately. The science is real. The cost reductions are real. The improved outcomes are real. The companies and institutions that organized around the information frame early are the ones now shaping the architecture.
Education does not have a Moderna yet.
It has districts, county offices, state agencies, trade associations, and governance-technology vendors who have spent the last decade producing better products. The next decade will not reward better products. It will reward the institution that builds the process — the measurement substrate, the early-detection screening, the differentiated intervention infrastructure, the agent-layer visibility for every child, the reinvented workday — and gets it approved at scale.
That institution exists in California. It has the statutory seat on the relevant working group, the distribution relationships to every district, and a thirty-year history of solving exactly the kinds of multi-stakeholder governance problems that the next decade will require solving. The decision in front of it is whether to step into the role medicine has demonstrated is possible — or to keep producing better products while another institution, in another state, builds the substrate first.
The mRNA moment arrived in medicine in 2020. It is arriving in education now.
The architecture that gets built in the next twelve months will set the trajectory for the decade. Biology is information. Education is information. The work of building the substrate is the work in front of us.
—