The Classroom That Became a Screen

Part I: A $163 Billion Industry After the Hangover

The EdTech industry experienced a classic boom-bust cycle. Global venture capital surged from $7 billion in 2019 to $20.8 billion in 2021, then cratered to just $2.4 billion in 2024 — the lowest level in a decade. By Q1 2025, quarterly investment had fallen to $410 million, down 35% from the same period the prior year. Over the past decade, roughly $80 billion in venture capital flowed into EdTech globally. Most of it did not produce durable companies.

The market itself continues to expand regardless. Precedence Research estimates the global EdTech market at $163.5 billion in 2024, growing to $348 billion by 2030 at a 13.3% compound annual rate. The U.S. market alone stood at $45.8 billion. The funding drought has triggered massive consolidation: Bain Capital acquired PowerSchool for $5.6 billion in June 2024, KKR acquired Instructure — the company behind Canvas — for $4.8 billion the following month, and the year saw 128 EdTech acquisitions total.

The most instructive cautionary tale belongs to 2U, which filed Chapter 11 bankruptcy in July 2024 carrying over $900 million in debt accumulated primarily through its acquisitions of edX and Trilogy Education. Once valued at billions, the online program management giant's collapse reverberated across higher education, affecting 260 university partners and casting doubt on an OPM model where companies typically took 60% of tuition revenue in exchange for managing online programs.

The survivors are thriving by different means. Duolingo reported $748 million in FY2024 revenue, up 41%, with 40.5 million daily active users and guidance for surpassing $1 billion in 2025. Coursera reached 191 million registered learners and posted $694.7 million in revenue. Stride Inc., the virtual charter school operator, hit $2.4 billion in FY2025 revenue with average enrollment of 234,000 students. And Chegg — once valued at $14 billion — saw its stock collapse 99% from its 2021 highs, losing over 500,000 subscribers since ChatGPT's launch. Chegg is the canary in the coal mine for what happens when AI replaces your core product overnight.

Part II: AI Tutoring — Promise Outrunning Proof

Khan Academy's AI tutor Khanmigo grew from 68,000 users in the 2023–24 school year to 700,000 in 2024–25 — a 731% increase — with expansion from 45 to over 380 school district partners. Sal Khan reported 1.4 million total users by April 2025. New Hampshire became the first state to sign a statewide contract, deploying Khanmigo across 50 districts serving approximately 40,000 students.

The uncomfortable truth: Khan Academy has not yet conducted a randomized controlled trial of Khanmigo. The organization's claim that it is 8 to 14 times more effective than independent learning comes from internal data, not independent verification. A Michigan Virtual pilot found that while teacher confidence grew, students frequently typed "idk" rather than engaging meaningfully with the AI tutor. The gap between deployment scale and evidence base is widening faster than the research can follow.

The most rigorous evidence comes from elsewhere. A Harvard University RCT published in Scientific Reports in June 2025 found that 194 undergraduate physics students using an AI tutor achieved learning gains more than double those of students in active-learning classrooms, with an effect size of 0.73 to 1.3 standard deviations. This is striking — but it was conducted at a single elite university, in a single course, compared to active learning rather than human tutoring. Generalizability remains uncertain.

A Stanford Tutor CoPilot RCT took a different approach, providing AI assistance to human tutors working with 1,800 K-12 students. Students whose tutors received AI support were 4 percentage points more likely to progress through assessments, and students of lower-rated tutors gained up to 9 percentage points in math proficiency — at a cost of roughly $20 per tutor per year. This human-AI hybrid model may be the most promising path forward: not AI replacing the teacher, but AI making the teacher more effective.

The cognitive costs of AI dependency are beginning to surface. An MIT Media Lab study tracked 54 participants over four months and found that ChatGPT users showed the lowest brain activity, lowest memory recall, and lowest self-reported ownership of their work. When LLM users were switched to no-tool conditions, they performed worse than participants who had never used AI — a pattern the researchers described as an accumulation of cognitive debt. A larger survey of 666 participants found a significant negative correlation between frequent AI tool usage and critical thinking abilities, with younger participants showing higher dependence and lower scores.

Student adoption is skyrocketing regardless. Pew Research reported in January 2025 that 26% of U.S. teens use ChatGPT for schoolwork, double the rate from 2023. The College Board found that 84% of high school students use generative AI for schoolwork, with 69% using ChatGPT specifically. RAND found 54% of K-12 students and 53% of teachers used AI for school in 2025. Simultaneously, 50% of students worried about being falsely accused of AI-assisted cheating — a concern validated by Stanford research showing AI detectors misclassified 61% of non-native English speaker essays as AI-written.

Teachers are adapting faster than institutional policy. A Gallup/Walton Family Foundation survey found 60% of K-12 public school teachers used AI during the 2024–25 school year. Teachers using AI weekly reported saving an average of 5.9 hours per week — equivalent to roughly six extra weeks over a school year. The most common applications: preparing to teach, making worksheets, and modifying materials for student needs. Yet 68% of teachers received no AI training during the school year.

Part III: A Literacy Crisis EdTech Has Not Solved

The 2024 NAEP results delivered devastating news. Only 31% of fourth graders scored at or above proficiency in reading, down 4 percentage points from 2019. Forty percent scored below the Basic level — the worst reading scores in more than 30 years for the lowest-performing students. By twelfth grade, only 35% were considered academically ready for entry-level college reading, the lowest in more than 20 years.

Reading scores have been on a downward trend for more than a decade, beginning well before COVID-19. The 2024 average fourth-grade score was not significantly different from the first NAEP reading assessment in 1992. The gap between higher-performing and lower-performing students has been widening; pandemic learning losses for the bottom 10% grew 70% larger between 2022 and 2024. Adult literacy tells a parallel story: the 2023 PIAAC results showed the U.S. average literacy score fell to 258, down 12 points from 2017, with approximately 130 million adults possessing low literacy skills.

The Science of Reading movement has generated enormous legislative momentum — 40 states plus D.C. have passed laws on evidence-based reading practices, and multiple states have banned the three-cueing system, where children guess words from context rather than decoding them phonetically. A 2019 EdWeek survey found 75% of K-2 teachers had been using three-cueing; 65% of education professors taught it. The scale of the pedagogical reversal is enormous.

EdTech's relationship to this reversal is decidedly mixed. Platforms like Lexia Core5 provide adaptive phonics-based instruction with strong evidence ratings. But many popular educational apps continue to use methods explicitly contradicted by the research — emphasizing sight-word memorization and guessing strategies over systematic phonics. A meta-analysis by Delgado et al. covering 54 studies and 171,055 participants found a consistent paper advantage over screens for reading comprehension, with the effect growing larger over time. Maryanne Wolf of UCLA argues we need a "biliterate brain" adapted to both mediums, but warns that the shallowing effect bleeds over into all reading contexts.

The Matthew Effect — the observation that strong readers get stronger while struggling readers fall further behind — operates with particular force in digital environments.

A longitudinal study using national data found two adverse effects of EdTech: a negative effect of excessive use on achievement scores, and a gap-widening effect where benefits were proportional to prior achievement. When Duke University researchers tracked broadband rollout across North Carolina, math and reading scores decreased in each region, with the negative impact greatest among economically disadvantaged students. A study of two Philadelphia libraries found that affluent toddlers received 17 times as much adult attention while using computers as their low-income counterparts. The technology is not the intervention. What humans do around the technology is the intervention.

Part IV: The Credential Revolution That Hasn't Arrived

Public confidence in the four-year degree is eroding. A Pew Research survey found only 25% of U.S. adults say a degree is extremely or very important for a well-paying job. Gallup reported in September 2025 that perceived importance of college hit an all-time low. Total student loan debt has reached $1.84 trillion across 42.8 million borrowers.

Enrollment numbers reflect this shift, though not as dramatically as the rhetoric. Fall 2024 undergraduate enrollment stood at 19.28 million, down 8.4% from the 2010 peak. But enrollment has actually been recovering — total postsecondary enrollment grew 3.2% in spring 2025 versus the prior year, and first-year enrollment jumped 5.5% in fall 2024.

The skills-based hiring movement has generated powerful corporate announcements. Indeed reported that 52% of U.S. job postings did not mention any formal education requirement. Thirty-one states pledged to remove degree requirements from state employment. Google treats its Career Certificates as equivalent to a four-year degree in its own hiring. IBM removed degree requirements from roughly 50% of U.S. roles.

But the most important study in this space tells a sobering story. Harvard Business School and the Burning Glass Institute found that while the number of roles dropping degree requirements nearly quadrupled between 2014 and 2023, actual hiring of non-degree candidates increased by only 3.5 percentage points. The net impact: fewer than 1 in 700 hires benefited from skills-based hiring reforms. Forty-five percent of firms were classified as "In Name Only" — they changed job postings but not hiring behavior.

Seventy million U.S. workers are what Opportunity@Work calls STARs — Skilled Through Alternative Routes. Over the past 20 years, STARs lost access to 7.5 million middle- and high-wage jobs; they have since regained only about 10% of that lost access. Micro-credentials are proliferating — Credential Engine's 2025 report counted 1.85 million unique credentials in the U.S. — but 46% of employers expressed quality concerns and 33% questioned credential validity. The New York Fed calculated in April 2025 that the college degree's ROI remains 12.5%, still beating long-term stock market returns. The degree is weakening as a signal, but it has not been replaced.

Part V: The Digital Divide Deepened When the Safety Net Collapsed

The Affordable Connectivity Program — a $14.2 billion initiative providing broadband discounts to 23 million households — ended June 1, 2024 when Congress failed to appropriate additional funding. An estimated 5 million households lost internet access entirely. A January 2025 survey found that 40% of former ACP households cut back on food to afford internet and 64% couldn't maintain regular contact with family members.

Roughly 5 million U.S. households with school-age children lack high-speed internet at home. Only 54% of households earning less than $30,000 subscribe to broadband, compared to 94% of the highest-income households. In rural America, 22.3% of residents lack terrestrial broadband coverage, compared to 1.5% in urban areas.

COVID learning loss persists with alarming stubbornness. As of spring 2024, the average student in grades 3–8 remained nearly half a grade level behind pre-pandemic levels in both math and reading. Math recovery has shown modest progress, especially in elementary grades. Reading recovery has stalled entirely — in third grade, the reading gap was unchanged between 2021 and 2025. High-income districts are almost four times more likely to have recovered than low-income districts. The gap between districts with the highest and lowest proportions of Black students grew by 17% from 2019 to 2024.

The $189.5 billion in ESSER pandemic relief funds — the largest one-time federal investment in schools in history — expired on September 30, 2024. Schools had invested heavily in technology, mental health, summer learning, and tutoring. When the cliff arrived, high-poverty districts faced roughly a 6% budget reduction versus 2% for low-poverty districts. Baltimore saw the immediate closure of afterschool programs for 3,000 students across 44 schools. Brookings had estimated in 2022 that schools would need $440 billion to fully recover lost learning for students in poverty. The $190 billion was never enough.

Part VI: The Homeschool Technology Ecosystem

Homeschooling has moved beyond a pandemic phenomenon into durable structural change. Approximately 3.4 million K-12 students were homeschooled during the 2024–25 school year, representing 6.26% of the school-age population — roughly double Catholic school enrollment. In 2024–25, homeschooling grew at an average rate of 4.9% year-over-year, nearly three times the pre-pandemic rate. Thirty-six percent of reporting states recorded their highest homeschool enrollment ever, exceeding even pandemic peaks.

The demographic profile has shifted. The homeschool population is now approximately 60% White, 19% Hispanic/Latino, and 12% Black — largely mirroring the general school-age population. Black homeschooling households rose from 1% to 8% between 2020 and 2021. The most commonly cited reason is concern about the environment of other schools, followed by desire for moral instruction and dissatisfaction with academic quality.

Technology is making homeschooling feasible for families who could never have considered it previously. Khanmigo costs $4 per month for families, with parents able to add up to 10 children. The combination of Khan Academy, Khanmigo, Time4Learning, and free tools like ChatGPT puts a credible K-12 education technology stack within reach for under $500 per year — versus the approximately $15,240 per student spent by public schools.

The microschool movement is the fastest-growing segment. The National Microschooling Network estimates 1.5 million K-12 students attend approximately 95,000 microschools — small learning environments using adaptive technology platforms with human guides rather than traditional teachers. Acton Academy has expanded to over 270 affiliates across 42 states and 26 countries. Students reportedly complete core academics in roughly 2 to 2.5 hours daily of focused, adaptive instruction.

Education Savings Accounts are the policy accelerant. Eighteen states have established ESA programs, with 11 offering universal or near-universal eligibility. Arizona's program has grown to 102,852 students with spending reaching $886 million in 2024–25. Florida's ecosystem serves over 500,000 students. Texas allocated $1 billion for Education Freedom Accounts in 2025. These programs create a direct public-funding pipeline for homeschool technology.

Part VII: Privacy, Surveillance, and the Teacher Crisis

In December 2024, PowerSchool suffered the largest breach of children's data in U.S. history, affecting approximately 62 million students and 9.5 million teachers. A 19-year-old attacker exploited compromised credentials and the absence of multi-factor authentication. Data exposed included names, Social Security numbers, medical alerts, and disciplinary records. Research has found that 96% of EdTech applications share student data with third parties. Districts utilize an average of 200 or more EdTech providers, creating nearly impossible compliance requirements. COPPA underwent its first major revision in over a decade in 2025, but FERPA has never resulted in funding being withheld from any institution in its 40-year history.

The proctoring industry expanded during COVID with documented bias: NIST found facial recognition performs considerably worse on darker skin tones, and Black students reported repeated failures to validate identity. UNESCO's 2023 Global Education Monitoring Report found that 89% of EdTech products recommended during the pandemic could surveil children. The school phone ban movement has accelerated from 24% of countries in June 2023 to 58% by March 2026. Finland, once the world's model for EdTech integration, has seen some schools return to physical books and pencils after years of issuing laptops to students as young as 11.

The teacher shortage compounds everything. The Learning Policy Institute estimated at least 411,500 positions were either unfilled or filled by teachers not fully certified — roughly 1 in 8 teaching positions nationally. Special education, science, and math are the most acute shortage areas. K-12 teachers have the highest burnout rate of any U.S. industry at 44%. Teachers earn 26.4% less than other professionals with similar education. Only 30% consider teaching rewarding; only 19% believe it is a sustainable career.

South Korea invested $760 million in AI-powered digital textbooks, launched in March 2025 — then watched adoption drop from 37% to 19% in a single semester before the incoming president pledged to rescind the policy entirely. China's 2021 Double Reduction policy banned for-profit K-12 tutoring, destroying companies valued at over $1.1 trillion combined, but demand simply went underground. Estonia, ranked first in Europe on PISA, achieved universal digital integration with 100% of schools using e-school platforms while maintaining smaller performance gaps between socioeconomic groups. The Estonian model demonstrates that technology and equity are not inherently in tension — but the technology must be deployed within a system that prioritizes teacher quality, universal access, and structured implementation.

Conclusion: The Dignity Question EdTech Must Answer

Three findings deserve particular emphasis. First, the Matthew Effect in EdTech is now well-documented across multiple studies and contexts: without deliberate design, implementation support, and human mediation, educational technology consistently benefits those who are already advantaged and widens gaps for everyone else. Second, the credential revolution is moving at the speed of press releases, not the speed of actual hiring — the Harvard finding that fewer than 1 in 700 hires have been affected by degree-requirement removal should temper any celebration of the death of the degree. Third, the most promising evidence consistently points not toward AI replacing human instruction but toward human-AI hybrid models — the Stanford Tutor CoPilot approach, where AI amplifies the effectiveness of human tutors at a marginal cost of $20 per year.

The 731% growth of AI tutoring is real. The 62 million student records exposed in a single breach are real. The 31% fourth-grade reading proficiency rate is real. The 5 million households that lost internet when the ACP expired are real. The 411,500 teaching positions that are unfilled or inadequately staffed are real. These are not contradictions. They are the same system, producing different outcomes for different people based on the same variables that have always determined who thrives in American education: income, geography, race, and whether someone with authority decided the investment was worth making.

The technology exists to narrow every one of these gaps. Whether it will depends entirely on choices that are not about the technology at all.