TikTok & AI Have Changed Education Forever - What it means for Teachers, Students & Parents
Summary
Teachers—not students—are emerging as AI education’s first strong customer base. Zach Cohen estimates MagicSchool has more than 5 million users and that roughly 50% of U.S. teachers have tried it, despite teachers’ limited software budgets. The value proposition is immediate: automate grading, feedback, assignments, and curriculum preparation so teachers can become “10 times better at their job” with less burnout.
Education has moved from AI prohibition to pragmatic procurement unusually quickly. After major districts banned generative AI, Zach now thinks about 80% of districts have teams evaluating it, while Claude for Education and OpenAI’s education platform are being partnered and piloted with universities; some schools, “I think Ohio State is one of them,” are making AI use mandatory. Higher education leads because AI literacy is increasingly viewed as preparation for work and everyday life.
Usage is measurable now, but learning efficacy remains years from a reliable benchmark. Zach favors monthly retention and cohort-level days used per week—hopefully above four or five—because exam-driven homework traffic is otherwise spiky. Annual tests require multiple years to isolate causation, and today’s studies showing gains from AI-instructed courses remain “research papers and case studies,” not statewide or nationwide evidence.
Alpha School is a high-end experimental signal, not yet a mass-market model. Its roughly $40,000 tuition, self-selecting families, large software budget, and freedom to experiment make it an education “labs team”; Zach says students ranked in the 99th percentile on some assessments and in the top 1–2% nationally. The investable question is whether falling software costs and early partnerships can make that kind of experimentation more accessible to resource-constrained public schools.
AI is unlikely to replace teachers soon because it has barely entered the instructional loop. Most current tools generate conventional worksheets and answer sheets rather than AI-native lessons in which students converse with historical characters, construct worlds, or turn writing into games. Zach expects active teaching to decline as AI improves, but his answer on full replacement is “no or very long horizon away.”
TikTok-style education reveals a larger opportunity: unbundle what is taught, how it is explained, and who delivers it. AI celebrity videos, NotebookLM podcasts, and Veo 3 historical vlogs let learners switch among visual, audio, reading, and practice modes by topic rather than accept a fixed “learner type.” Yet distribution remains the bottleneck: textbook publishers still gatekeep classroom content, while schools overwhelmingly use productivity tools instead of the more engaging AI-native experiences.
Parent adoption will follow provable outcomes and sharp economic comparisons, not AI enthusiasm. Zach cites an early reading product promising to bring three- and four-year-olds to third-grade reading level within three months for $500 a month: success could unlock substantial demand, while failure sends families back to tutors. AI competes poorly with a $100–$200-an-hour tutor for families that can afford one, but compellingly with four hours of Netflix for a family weighing screen time.
Deep dive
1. Education has crossed from AI hysteria into procurement
Zach describes the first phase as an “AI wave and AI detectors” followed by the “Rock ’Em Sock ’Em Robots of education”: Los Angeles and New York City public schools banned the technology almost immediately. Eighteen months later, he believes the sector has passed the “hysteria moment” and entered a “pragmatic moment”—fast by education standards, where even moving to cloud systems has taken five or six years and remains incomplete.
K–12 retains more skepticism, especially around whether teachers should see and supervise students’ AI interactions. Even so, Zach thinks roughly 80% of districts now have generative-AI teams, earmarked budgets, and staff actively evaluating products; his net assessment is that teacher oversight is healthy and “there’ll be more AI in the classroom.”
Higher education is moving faster: Claude for Education and OpenAI’s education platform are being piloted with universities as horizontal infrastructure for both students and faculty. Zach’s explanation is practical rather than ideological—people increasingly recognize that they “are going to need to know how to use” AI in jobs and daily life.
The market’s fragmentation remains fundamental. Public, private, charter, homeschool, parent-paid supplements, universities, adult learning, and corporate reskilling each have different buyers and adoption paths, so “education” is not one go-to-market category despite sharing the same underlying technology shift.
2. Teacher productivity is the first proven wedge
Zach’s surprise is that teachers, not students or adult learners, show the strongest willingness to pay and incorporate AI into daily workflows. Students often use homework helpers just long enough to finish a problem set; existing platforms such as Duolingo may improve completion or efficacy with AI, but he has not yet seen a native-AI adult-education winner, perhaps because distribution and established pedagogy matter more than a new interface.
Teachers feel immediate ROI because “90% of the job that they hate is the administrative part”: grading, feedback, building assignments, and updating curricula after hours. Instead of borrowing the same materials year after year or revising one unit at a time, they can generate curricula per student—even if today’s outputs are still mostly worksheets and multiple-choice questions.
MagicSchool reportedly has more than 5 million users, and Zach thinks about half of U.S. teachers have used it. That makes teacher-led, bottom-up sales the most commercially mature AI-education motion so far: $15 or $20 a month is meaningful from a small teacher wallet, but the productivity return can be “outsized.”
For investors, Zach separates engagement from educational proof. He watches monthly retention and cohort-level days used per week—preferably above four or five—to distinguish a learning habit from deadline-driven traffic; simplistic weekly or yearly retention can be a “red herring” because summers, exams, and two-day test cramming distort the curve.
3. Learning outcomes remain unproven at system scale
The hard question is what “working” means. Product retention can show that learners return, but annual assessments require several years of results and careful separation of dependent and independent variables; meanwhile, AI remains “at the periphery of education,” helping teachers create familiar materials and students get homework done rather than changing the core schooling experience.
Zach has seen university and school research reporting test improvements from AI-instructed courses, which he calls exciting. His hedge is load-bearing: these are still “research papers and case studies,” not statewide or nationwide studies, so founder fluency in pedagogy can help supplement missing longitudinal evidence.
Alpha School offers the clearest experimental signal. Zach likens it to a corporate labs team that can rapidly test features, discard failures, and keep what works: roughly $40,000 tuition, ample software spending, and families deliberately choosing technology remove friction that would constrain a public district.
The reported results are striking but conditional—around the 99th percentile on several assessments and top 1–2% nationally. Zach sees this as evidence that going “full tilt on AI and education” can have outsized impact, while acknowledging that integration costs, software literacy, commercialization, and public-school budgets still complicate replication. He also views Alpha’s emphasis on self-discovery, self-learning, and limited instruction as part of what is working.
4. AI-native instruction is more radical than better worksheets
Justine presses the controversial budget question: if superintendents cannot spend hundreds of thousands on software but already spend heavily on teachers, does software eventually replace them? Zach’s answer is “no or very long horizon away”: shortages persist, teachers are overloaded, and classroom AI lags consumer and enterprise adoption by multiple years.
The distinction is between improving a teacher’s workflow and changing a student’s environment. A freshly generated worksheet with newer memes remains the same educational asset; an AI-native unit might let students debate history with an avatar, create a world from a writing prompt, or build a game around it. “We’re so far away from even AI teaching units,” Zach argues, that AI teachers are farther still.
He does expect the amount of active teaching to fall because AI can eventually perform that function well, but he does not expect full human replacement. The near-term product is augmentation: teachers become “10 times better at their job,” burnout falls, and students only gradually begin interacting with AI as the instructor itself.
5. TikTok formats unbundle content, delivery, and distribution
Free AI education is evolving beyond Khan Academy-style videos into NotebookLM textbook podcasts, Unlock Learning clips, deepfake celebrity explanations for AP or IB subjects, and Veo 3 “day in the life” historical vlogs. Zach says the clips “feel like brain rot content, but they’re actually the complete opposite”: their pacing and aesthetics are familiar, yet the underlying subjects can be detailed and technical.
The best specimen is the evolution from Taylor Swift explaining Taylor series to Drake and Sydney Sweeney discussing the length of a 3D vector. Better animation, graphics, and synthetic delivery separate the quality of the explanation from the identity presenting it—allowing both layers to be optimized independently.
That breaks the old habit of labeling someone a visual, audio, or reading learner. Zach may want visuals for one subject, a podcast for another, text for a third, and 100 practice problems for a fourth; modality can now vary with the topic, the learner’s current understanding, and whether the goal is casual fluency or a consequential exam.
Adoption still favors familiar tools. Companies enter schools through worksheet generators, student helpers, and feedback products, but usage of their more imaginative conversational experiences remains “extremely low.” Olivia says there will probably need to be a substantial professional-development movement around bringing AI “not just into your workflow, but into your classroom.”
6. Outcomes, gatekeepers, and engagement will decide the next market
Justine says textbook publishers still control much of what enters classrooms. They can treat AI as cannibalization or as an extension of proprietary source material; she sees the next 18 months as depending on where publishers invest in net-new products. They may partner with AI companies—trading distribution for stronger products—or watch existing content “slowly losing its value day by day.” Olivia says publishers need to grow innovation arms fast or outsource them.
Parents’ reward model is simpler: “Parents want better outcomes.” Zach’s early reading-company example charged $500 monthly and promised three- and four-year-olds third-grade reading ability within three months; parents signed up quickly, but future demand depends on proving that promise rather than merely offering AI.
Adoption will be socioeconomic, geographic, and resource-dependent. AI has a harder comparison against a tutor costing $100–$200 an hour, but a much easier one against four hours of Netflix; configurable LLMs could add appeal by letting parents restrict topics or require every answer to show an equation, proof, and explanation.
Over the next 12 months, Zach expects higher education to lead, voice and real-time interaction to move AI inside classrooms, and the market to learn whether foundation-model companies are sufficient or need specialized applications above them. Olivia thinks education will still look fairly similar; Zach says it may not produce a massive company yet: “I think we’ll learn a lot as the market kind of matures.”
Justine’s desired endpoint is a fully AI-native teacher influencer—not a celebrity deepfake, but an adaptive character optimized for engagement, explanation, and each student’s pace. Zach agrees the models are “good enough to teach,” with verification still necessary; he says LMSs are now trusted as learning partners, and the missing layer is an engaging experience for moments when learning is mandatory, difficult, and stressful.