When a new technology arrives, people almost always make the same mistake: they use it to recreate whatever came before.
The first cars looked like carriages without horses. Early movies looked like filmed stage plays, with stationary cameras, painted scenery, and actors entering and exiting as though they were still onstage. Early television borrowed heavily from radio before people understood what television itself could become.
The first websites were brochures placed on a screen. Before search engines became dominant, people created directories of websites, imagining that finding information online would work something like looking up a business in a telephone book (remember those?)
Eventually, someone stops asking, “How can this technology improve the existing system?” and asks a more intriguing question: “What kind of system would we build if this technology had existed from the beginning?” That is how we eventually got social media—for all the good and bad that came with it—rather than simply better online directories and text articles.
Education has been through this cycle before. Early online learning often consisted of printed materials converted into PDFs and placed on a website. The content moved online, but the educational model barely changed. During COVID, many traditional schools repeated the pattern. The result was exhausting for teachers, frustrating for students, and mistakenly called “online learning.” It was the physical classroom delivered through a screen, not a good online course.
It’s happening again with artificial intelligence. Schools are mostly using AI to write lesson plans, generate quizzes, create worksheets, tutor students, and help teachers grade assignments. These applications are useful, but they mostly make the existing system more efficient and save teachers some time.
In physical schools, students remain divided primarily by age, moving through subjects on fixed schedules. Twenty to thirty students are expected to learn similar material at roughly the same pace. Teachers spend enormous amounts of time delivering information, evaluating work, and documenting progress. Learning remains organized around courses, semesters, and credit hours because those were the practical structures of an earlier era.
Online schools used to be a focal point of innovation in K–12 education, but they are also mostly treating AI as a way to improve their existing models rather than as a reason to rethink it.
Somebody—perhaps a school, a network of schools, or an organization that does not yet exist—is going to build an educational model that is native to AI. It might truly organize learning around individual progress rather than age and seat time. It might give each student continuous instruction, feedback, and support while teachers concentrate on relationships, judgment, motivation, discussion, and the deeply human parts of education.
These ideas aren’t new. Online learning advocates have promised many of these benefits for years, and the results have often fallen short. But AI changes the technology enough that the promise is worth reconsidering.
Where might this new model emerge?
Alpha School and the broader 2 Hour Learning model are one answer. Students are supposed to complete their core academic work in no more than two hours a day through personalized, software-driven instruction. The rest of the day is devoted to workshops, projects, physical activity, and life skills, with adult “guides” serving as coaches and motivators rather than traditional classroom teachers.
Alpha is also moving beyond the experimental stage. Axios recently reported that it plans to grow from roughly a dozen campuses to about 50 during the 2026–27 school year, entering 27 markets. Many of the schools remain small, although its New York City campus is expected to grow from 22 students to 160. Alpha says its guides are paid at least $100,000 annually, which is another interesting element of the model: the idea isn’t simply to replace teachers with cheap technology, but to change what the adults do.
Whatever one thinks of Alpha, it is much closer to reconsidering school around new technology than simply using AI to generate worksheets and lesson plans.
But Alpha also raises an important question about how innovation spreads. Tuition starts around $40,000, with Axios reporting prices as high as $75,000. That makes Alpha a very expensive laboratory, and it is hard to see how the current model translates directly to public-school funding or to a much broader and less self-selected group of students.
None of that means Alpha is headed in the wrong direction. But it suggests one possible route for AI-native education: develop the model first with families willing and able to pay a premium, learn what works, and eventually figure out whether it can be made cheaper and more broadly available.
There is another possible route, and it looks almost exactly opposite.
Clayton Christensen argued that disruptive technologies often take hold first among “nonconsumers”—people whose alternative is not the established product, but nothing at all. Early online learning followed this pattern remarkably well. It served students whose schools could not offer an AP class, a specialized course, credit recovery, or enough academic breadth, particularly in small and rural schools. Homebound and homeschooled students were other early areas of nonconsumption. Florida Virtual School grew in part because physical schools needed additional capacity and ways to address class-size pressures.
In other words, online learning didn’t initially have to persuade everyone that it was better than a traditional classroom. It could start by providing something students otherwise could not get.
Where is the equivalent opportunity for AI?
That is less obvious. The education landscape is very different than it was in the 1990s. Many students now have access to online schools, hybrid schools, microschools, homeschooling options, dual enrollment, online courses, and a huge universe of information and instruction outside school. There are fewer obvious empty spaces.
But there are still plenty of unmet needs. Many students cannot get an individual tutor, a particular course, flexible pacing, immediate feedback, a subject-matter expert, or a workable path to recover credits. Far too many students are still not completing high school, and among those that do, too many are unprepared for their next steps.
The catch is that many students with the fewest options are also the students most likely to need substantial human structure, motivation, and support. Giving them better instruction may be only part of what they need.
So perhaps Alpha and other high-end private schools will be the laboratories where AI-native education is developed. Possibly the breakthrough will instead come from solving problems for students who cannot get what they need from the existing system. Or perhaps today’s online schools will figure out how to build the new model themselves.
The most important educational use of AI will not be a better worksheet, a faster lesson plan, or a chatbot added to the learning management system. It will be the educational system that could not have existed without it.
Can today’s schools build that system? We will find out.
************************
PS: After I’d drafted this post but before publication, this post from Dan Meyer came out, digging into Alpha School further. It’s worth a read for anyone interested in AI in education.