AI for Education in 2026: Tools, Schools, and Methods

An 11-year-old in Austin spent two hours on her laptop this morning. Math first. Then reading. Then a science topic she’d picked the day before. By 10:30am she had covered what most public-school sixth-graders take five hours to cover. At 11am she headed to a workshop where she and seven other kids built a working model bridge out of basswood. After lunch she had violin. Then chess club. Then a creative writing project she’d been working on for a month. She tests in the top 1-2% of the country on the standardized assessments her school administers. She also says, when adults ask, that she “loves school.” Her tuition runs in the high tens of thousands. Her school is called Alpha. Most American parents have never heard of it. The technology that makes her two-hour academic day possible is mostly not what people mean when they say “AI” — it’s adaptive software that’s been around in some form for fifteen years, layered with a few newer AI-tutoring components and a human “guide” whose job is mostly motivational. The thing Alpha is selling isn’t the software. It’s the model.

That model is replicable, partially, by parents who do not have $50,000 a year and never will. That replication is the most important thing happening in education in 2026, and almost no mainstream education writing covers it well. This guide is my attempt to fix that for parents, homeschoolers, adult learners, and anyone who walked into a conversation about AI and learning and walked out more confused than when they walked in. If you read nothing else here, read the part on Bloom’s 2-sigma problem and the part on what AI doesn’t replace. Both are load-bearing for the rest.

What’s actually happening to American education right now?

Before the case for AI in learning, the case for changing what we’re doing. The numbers from the National Assessment of Educational Progress — the “Nation’s Report Card,” which has been measuring American schoolchildren since 1969 — describe an emergency that most parents are aware of in the abstract and underestimate in the specifics.

Measure	Latest finding	Source
4th-grade reading proficiency	Only ~31% scored “proficient” or above on the 2024 NAEP — the lowest in over thirty years	NAEP, National Center for Education Statistics
8th-grade reading proficiency	~30% proficient; reading scores at 33-year lows	NAEP 2024
8th-grade math proficiency	~28% proficient; lowest score in NAEP’s history	NAEP 2024
13-year-olds’ reading scores	Fell to lowest level since 1971 in the Long-Term Trend	NAEP Long-Term Trend Assessment 2023
COVID learning loss	Roughly half a school year of math achievement lost; not yet recovered four years later	Eric Hanushek (Stanford) / Harvard Center for Education Policy Research
U.S. PISA math ranking	~28th globally; below Slovenia, Vietnam, Latvia	OECD PISA 2022
Adult literacy	~21% of U.S. adults read at “below basic” literacy	OECD PIAAC (Programme for the International Assessment of Adult Competencies)
Chronic absenteeism (K-12)	Roughly doubled post-pandemic to ~30% of students nationally	U.S. Department of Education, AEI
ACT composite score	Six consecutive years of declines; 2024 average was the lowest in 30+ years	ACT, Inc.
Teacher shortage	Estimated ~55,000 unfilled K-12 teaching positions; ~270,000 underqualified	Annenberg Institute / Brown University

The decline did not start with COVID. NAEP scores were drifting down through the late 2010s before the pandemic. COVID accelerated a curve that was already bending the wrong direction. The reasons are complex — chronic absenteeism, screen time, the “Sold a Story” reading-instruction problem documented by journalist Emily Hanford (American Public Media) where a generation of children were taught reading using a discredited “cueing” approach instead of phonics, post-pandemic disengagement, teacher attrition — but the direction is unambiguous.

This context matters because it changes the question. The question is not “is AI better than the school my child currently attends?” The question is “given what my child is currently getting, what’s the best alternative or supplement I can construct?” For a meaningful share of American children in 2026, even an imperfect AI-augmented learning setup is a measurable upgrade over what they’re currently receiving. That is a deeply uncomfortable sentence to write. It also matches what the data shows.

Why does any of this matter — what’s actually new in 2026?

In 1984, the educational psychologist Benjamin Bloom published a now-famous paper titled “The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring.” His finding, in plain English: a child who works one-on-one with a human tutor performs roughly two standard deviations better than a child in a traditional classroom. Two standard deviations is enormous — it’s the difference between a child who scores at the 50th percentile and a child who scores at the 98th. Bloom challenged the education-research community to find a classroom method that could match one-on-one tutoring. Forty years later, nobody has.

The 2-sigma problem matters because it tells you what’s possible. Children who learn 1:1 perform astonishingly better than children who learn in groups of 25. The reason no school district has switched to 1:1 tutoring is not that they don’t want to — it’s that one human tutor per child is unaffordable at any scale. Until 2024 or so, there was no realistic alternative. In 2026 there are several. Large language models like ChatGPT and Claude, plus adaptive learning systems like the ones Alpha School uses, can plausibly approximate 1:1 tutoring at one to two orders of magnitude lower cost than human tutors. This is the first time in human history that the 2-sigma effect could be available to a typical family.

What’s actually new in 2026 is not that the AI tools exist. They’ve existed for two years. What’s new is that the playbooks for how to use them — what Alpha School calls 2 Hour Learning, what MathAcademy.com calls knowledge-graph mastery, what the Khan Academy Khanmigo team is iterating on every quarter — have matured enough to produce repeatable results. The early-adopter phase ended. The middle-adopter phase is starting.

What are the three stories in AI for education that actually matter?

Most of the noise in this space is hype. Three threads under the noise are actually consequential.

One: school replacement. A small but growing set of schools have decided that the traditional classroom is the wrong unit of education altogether. Alpha School in Austin, Texas, is the lead case. Children spend ~2 hours/day on academic content delivered by software (the 2 Hour Learning model), then the rest of the day on workshops, projects, sports, life skills, and what Alpha calls “level-up activities.” Synthesis, founded by Josh Dahn and originally based in Tesla’s child-school, is another model — heavy on group problem-solving games rather than software-only academics. These schools cost $40,000-$75,000 per year in tuition, which is why most people haven’t heard of them. The interesting question for everyone else is whether the model can be approximated at far lower cost. The short answer is yes, partially, and the longer answer is in the next post in this cluster.

Two: AI-powered mastery-learning platforms. “Mastery learning” is an old educational idea — also pioneered by Bloom — that says a child should not move on to the next topic until they’ve mastered the current one. Traditional schools can’t do this because they have to keep the whole class on a pace. AI-powered platforms can. MathAcademy.com, run by Justin Skycak and the team that originated the Eurisko program at Pasadena High School, builds a knowledge graph for math and walks students through it problem-by-problem at exactly the pace their accuracy data supports. Khan Academy’s Khanmigo layer does similar work across math, science, reading, and writing, with an LLM tutor on top. These platforms are not magical, and they are not a complete education — but on their narrow turf (math practice especially), they outperform anything humans have built before.

Three: AI as homeschool partner. The U.S. homeschool population is about 3.1 million children (NCES estimates plus state-level filings, post-2020 surge). For these families, AI is not a curiosity — it’s already part of daily learning, used as tutor, content generator, lab partner, language coach, and writing critic. The interesting story isn’t whether homeschoolers use AI (they do). It’s how AI integrates with each major homeschool philosophy — Classical, Charlotte Mason, Montessori, Waldorf, Unschooling, Moore Formula, Thomas Jefferson Education, and the dozen others. Each philosophy has a different answer to the question “what is education for?” — and therefore a different answer to “how should AI fit in?” We’re building a full homeschool hub covering each.

What does AI actually do well in learning right now?

Task	How well AI does it (mid-2026)	Best tools
Answer subject questions at a tutor’s level	Strong — better than most non-specialist humans	ChatGPT, Claude, Gemini, Khanmigo
Math practice with adaptive difficulty	Strong — outperforms human-graded homework	MathAcademy, Khan Academy, IXL, DreamBox, ALEKS
Language conversation practice	Strong — patient, infinite-rep partner	Speak, Duolingo Max, ChatGPT Voice
Reading comprehension support	Strong — can summarize, explain, quiz	NotebookLM, Claude, Khanmigo
Writing feedback (mechanics + structure)	Strong	Claude, ChatGPT, Grammarly
Subject explanation in different ways	Strong — re-explain till it clicks	Any LLM
Assessment / diagnostic tests	Strong for narrow domains; weak for cross-disciplinary	MathAcademy, ALEKS, Khan Academy
Generating practice problems	Strong	Any LLM + the platform-specific tools
Vocabulary acquisition (foreign + native)	Strong — spaced repetition is solved	Anki + LLM, Membean, Quizlet Q-Chat
Public speaking / fluency coaching	Moderate-to-strong (newer category)	Orai, Yoodli, ELSA Speak

What does AI fail at — or what hasn’t changed since the printing press?

This section is the load-bearing one. The whole “AI changes everything” narrative falls apart if you actually look at what AI cannot do for a learner. The list is long.

Physical books in a child’s hand still beat screens for deep reading. Anne Mangen’s research in International Journal of Educational Research and Maryanne Wolf’s Reader, Come Home (HarperCollins, 2018) both document that long-form comprehension and recall are measurably higher when readers engage with paper than with screens. The reasons are partly haptic (touching the physical page anchors memory in space and time), partly cognitive (screens prime skimming rather than deep linear attention), and partly environmental (a book has no notifications). A child surrounded by physical books reads differently than a child surrounded by tablets. This isn’t a minor point. We have an entire dedicated post on the research because it matters more than every AI tool combined for learners under 12.

Handwriting strengthens memory in ways typing does not. Mueller & Oppenheimer’s 2014 study in Psychological Science, “The Pen Is Mightier Than the Keyboard,” found that students taking handwritten notes performed better on conceptual questions than students typing notes — even when the typists captured more verbatim content. The bilateral motor activation of cursive specifically (left- and right-brain coordination) shows up in fMRI studies as more pervasive cortical engagement than typing. A child who only types is missing a developmental input. AI tutoring delivered exclusively through a screen, with no parallel handwriting practice, gives up part of the cognitive scaffold humans have always used to learn.

Struggle is a feature, not a bug. Robert Bjork’s “desirable difficulties” research at UCLA documents that information learned with friction is retained longer and applied more flexibly than information learned smoothly. Henry Roediger and Mark McDaniel’s Make It Stick (Harvard, 2014) synthesizes decades of testing-effect research showing that retrieval practice — being forced to recall information without help — outperforms re-reading by a wide margin. John Sweller’s cognitive-load theory frames the same idea from another angle: learning happens when working memory is taxed productively, not when it’s offloaded entirely. AI tutors that always give the answer when the student is stuck produce a confidence that doesn’t survive the test. Tutors — human or AI — that withhold the answer at the right moment, force the rephrasing, push the student to articulate why they’re stuck, produce durable learning. Most AI tutors default to the wrong setting. The good ones can be configured to push instead of soothe.

In-person discussion is not replaceable. Mortimer Adler’s work on the Great Books, the seminar method that drives St. John’s College, the Socratic dialogue that anchors classical Christian education — these are pedagogies built around a teacher and a small group of students arguing about ideas. AI cannot replicate the felt accountability of arguing in front of three peers and an adult who knows you. Both teach the same content. They do not teach the same skill. A child who has only AI tutoring has missed an important part of what education does to a person.

Source-of-truth and durability are real concerns. Studies from Harvard Law and Columbia have shown that roughly 50% of URLs cited in published court opinions and academic articles are dead within 7-10 years. Wikipedia gets silently re-edited. Models hallucinate. Cloud storage gets price-hiked or shut down. If a family relies on a digital-only learning record, a digital-only family archive, a digital-only library — they are one platform decision away from losing all of it. We cover the practical implications in the physical-archive and source-of-truth post.

AI cannot replace a person who knows and loves the student. The variable that predicts learner outcomes more reliably than any other is whether a child has at least one adult who is paying close attention to them. AI does not provide this. AI’s role is to free up the adults’ time so they can provide it. If you spend the freed-up time on the adult’s phone, the AI added nothing.

What’s the practical path for a parent in 2026 — what should you actually do?

The honest answer depends on whether your kid is in public school, private school, homeschool, or you’re trying to make a decision. Here’s a thirty-day starter that works in all four cases.

• Week 1 — Diagnostic. Pick one subject your child is weak in. Sign up for a free trial of one mastery-learning tool that covers it — MathAcademy or Khan Academy for math, ALEKS for math+chemistry, Beast Academy for elementary math, Membean for vocabulary, Speak for foreign language. Let your child use it for 20 minutes a day. Watch what happens.
• Week 2 — Stack with an LLM. Add a separate 15-minute session a few times that week where you sit next to your child and use ChatGPT or Claude together on something they’re curious about — anything. The goal is for your child to see an adult use AI well so they learn what good use looks like.
• Week 3 — Add a physical book and a notebook. Pick one book at their reading level. Get a paper notebook. The rule for week three: every AI session has a follow-up where they write something by hand about what they learned. Two sentences is fine. The point is the motor act of writing, which engages brain networks that typing skips entirely (see our deep-dive: Why writing by hand activates more of your brain than typing).
• Week 4 — Evaluate. Did they progress on the diagnostic tool faster than they were progressing in school on the same topic? Did they like it? Did their handwriting-recall improve? Adjust the stack. Drop tools that didn’t earn their keep. Add one new tool you’ve been curious about.

This isn’t a curriculum. It’s a posture: try things, measure them, keep what works, drop what doesn’t, refuse to be tied to anyone’s marketing language.

What’s the practical path for an adult learner?

Adult learners — career switchers, late college students, GED candidates, people who never finished a degree, retirees curious about a topic — have a different problem than children. They have less time and more agency. The right starter:

• Pick a domain. Programming. Statistics. Spanish. Music theory. American history. Whatever you’d take if there were no cost barrier and no schedule constraint.
• Find the canonical resource. For programming, that’s still usually a real textbook or a course on a structured platform. For statistics, MathAcademy or a real textbook. For language, Anki + an LLM voice partner + a real textbook. The canonical resource gives you the spine.
• Use AI as the patient explainer. Every time you don’t understand a paragraph in the canonical resource, paste it into Claude or ChatGPT and ask for three different framings. Ask for an analogy. Ask for a simpler example. Ask it to quiz you.
• Write by hand at the end of each session. One paragraph about what you learned. This is the consolidation step that everyone skips and that determines whether you actually remember anything.
• Find one human. Even one. A friend, a forum, a Discord, a study group, a tutor for the hard parts. The accountability matters more than the explanations.

An adult learner using this stack at one focused hour a day will outpace the version of themselves who took a community-college course on the same topic, costs less, and retains more. I’ve watched this work for friends three times in the last eighteen months.

Which AI learning tools are worth the time to try?

Category	What it does best	Tools to know	Pricing as of 2026
Mastery-learning math	Adaptive problem sequences with knowledge graphs	MathAcademy.com, Khan Academy + Khanmigo, ALEKS, Beast Academy, IXL, DreamBox	$0 (Khan free) to $49/mo (MathAcademy)
Vocabulary + spelling	Spaced repetition + context	Membean, Quizlet Q-Chat, Vocabulary.com, Anki + LLM	$0 to $30/mo
Foreign language	Conversation practice + spaced rep	Speak, Duolingo Max, ChatGPT Voice, Pimsleur, LingQ	$0 to $30/mo
Reading + comprehension	Summarize, quiz, dialogue with text	NotebookLM, Claude with PDFs, Khanmigo, Beast Academy	$0 to $20/mo
Writing	Mechanics, structure, feedback at draft level	Claude, ChatGPT, Grammarly, ProWritingAid	$0 to $30/mo
Public speaking + fluency	Real-time AI coaching	Orai, Yoodli, ELSA Speak	$10-$30/mo
Science labs + chemistry	Concept explanations + practice	ALEKS Chemistry, Khanmigo, Brilliant	$0 to $25/mo
Computer science + coding	Practice problems, code review, conceptual help	Brilliant, CodeAcademy, Claude/ChatGPT, Replit	$0 to $30/mo
Early literacy (K-2)	Phonics, sight words, early reading	Hooked on Phonics, Reading Eggs, Khan Kids, Lalilo	$0 to $20/mo
General AI tutor	Patient explainer, brainstorm partner	ChatGPT, Claude, Gemini	$0 to $20/mo
Niche-specialty tools	Specific subjects, accents, styles	TeachTales, Twin Pic, FastMath, ChatABC, Egumpp, MobyMax	Varies by tool

I have dedicated post-by-post deep dives planned for each of these tools — we will publish them through the rest of 2026. The pillar guide stays general so it doesn’t go stale every time a tool changes its pricing.

What schools and learning models should you know about?

Model	What it is	Cost	Why it matters
Alpha School (Austin + expanding)	2 Hour Learning model; software-led academics + workshops	$40K-$75K/yr tuition	The proof point that the model produces top-percentile results
Acton Academy (multi-site, hybrid franchise)	Self-paced learner-driven model; older lineage	$10K-$30K/yr depending on campus	The micro-school precursor to Alpha
Synthesis	Group problem-solving sessions; originally Tesla-school spinoff	$200/mo for current Synthesis Tutor app	A different replication of the Alpha thesis
Microschools (general)	Small in-person school, often hybrid AI + human guide	$10K-$30K/yr typical	Fastest-growing K-12 segment in the U.S.
Homeschooling (traditional)	Parent-directed; many sub-philosophies (Classical, Charlotte Mason, Waldorf, Montessori, Unschooling, etc.)	$500-$5,000/yr per family on materials	About 3.1M U.S. children; AI is integrating fast
Hybrid / University-Model (UMS)	Half-week in-school, half-week at home; AI fills gaps at home	$5K-$15K/yr	Growing as parents balance work + presence
Worldschooling / Roadschooling	Travel-based; AI carries the curriculum	Highly variable	AI made this practically viable for the first time

The lesson under this table: the institutional answer to “what is a school” is fragmenting. Twenty years ago, almost all U.S. K-12 children were in public, parochial, or traditional private schools. In 2026, multiple credible alternatives exist at multiple price points. AI is the technology that made many of them viable.

What are the homeschool philosophies and how does AI fit each?

The single biggest mistake new homeschoolers make is treating “homeschool” as one thing. It isn’t. There are at least a dozen distinct philosophies, each with a different theory of what education is for, and each fits AI differently. The full guide is in the homeschool hub; here’s the one-paragraph version of each.

• Classical: Built on Dorothy Sayers’ 1947 essay “The Lost Tools of Learning” and the trivium (grammar, logic, rhetoric). AI fits the logic and rhetoric stages well; the grammar stage benefits from physical-book and memory-work emphasis.
• Charlotte Mason: “Living books,” narration, copywork, nature journals, short lessons. AI is a careful add-on — Mason philosophy is wary of “twaddle.” Use AI for the parent’s prep more than the child’s input.
• Montessori: Prepared environment, self-directed work, mixed ages, hands-on materials. AI fits older Montessori students well; younger Montessori (3-9) should keep AI minimal.
• Waldorf (Steiner): Delayed academics, screen-skeptical, story-and-craft-based. The most AI-skeptical of mainstream philosophies. Limited fit — and that’s a feature, not a bug.
• Unschooling: Interest-led, child-directed, no formal curriculum. AI is a perfect tool for the unschooler — infinite patient resource for whatever rabbit-hole the child is in this week.
• Moore Formula (Raymond & Dorothy Moore): Delay formal academics, emphasize work, service, and family. AI use trails behind life experience here.
• Thomas Jefferson Education (TJEd): Phases of learning; mentor-driven. AI as the mentor’s research staff.
• Wild + Free / nature-based: Outdoor-first, seasonal rhythm. AI is the parent’s research aid, not the child’s.
• Project-based / Acton-style: Real-world projects, learner agency. AI is core; this is the Alpha-adjacent approach.
• Eclectic: The most common in practice. Pull from many philosophies. AI fits because eclectic homeschool always fits whatever works.
• Robinson Curriculum: Self-teaching from a fixed set of historic books. Limited AI use by design.
• Worldschool / Roadschool: AI is the portable curriculum.
• Christian curricula: A.C.E., Abeka, Sonlight, BJU, Memoria Press, etc. Each handles AI differently; varies by publisher.
• Secular curricula: Oak Meadow, Build Your Library, Blossom & Root, etc. Generally AI-friendly.

Each of these will get its own dedicated post over the next several months. The work is to write each in the philosophy’s own voice — not to impose a one-size-fits-all “AI integration” template.

What’s coming next — what trends should you watch?

• The 2 Hour School Day becomes a category, not a single school. Alpha is the first. Synthesis is differentiated. There will be a dozen credible offerings by 2028, at multiple price points. Pricing will fall.
• Public-school adoption of AI tutoring as a parallel system. Several state legislatures (Florida, Texas, Arizona) are funding AI-tutor pilots in public schools. The next two budget cycles tell us whether this becomes structural.
• The microschool boom continues. Networks like Acton, Prenda, KaiPod, Wonderschool are scaling. Parents who don’t want full-homeschool but don’t want public school find these.
• AI tutoring meets the homeschool co-op. Parents take turns running mixed-AI / mixed-human lessons. Expect this to become the dominant micro-community structure.
• Standardized assessment lags. Standardized tests still measure what classroom teaching teaches. AI-trained students sometimes underperform on tests not because they don’t know the material but because the tests measure something orthogonal. This is a real friction point.
• College admissions evolve slowly. Some colleges are starting to take Alpha-style portfolios. Most are not. The transcript problem remains real for families on these paths.
• The “AI cognitive offload” debate intensifies. Will using AI weaken kids’ independent thinking? Mixed early evidence, depends entirely on how AI is used. Lazy use weakens. Active use doesn’t.

The Beginners in AI position on AI and learning

This is the most consequential moment for American education in living memory. Used well, AI tutoring lets a kid skip the wasted instructional minutes most classrooms still produce and lets an adult reach a level of literacy they were never going to find on their own. We are here for both.

The companion principle is the one most takes miss. The technology is at its best when it enhances what a learner does, not when it replaces the parts that build the learner. Handwriting builds neural connections typing cannot. Reading a whole book builds a kind of attention skimming and summaries cannot. Sitting with a hard problem builds the muscle that asking an LLM to solve it skips.

Use AI. Use a pen. Read books. Watch your kid struggle for ten minutes before you intervene. That is the stance behind every recommendation on this page.

Frequently asked questions

Should I let my elementary-aged child use AI directly?

Not unsupervised, and not as a substitute for parent or teacher attention. For elementary children (~K-5), the right pattern is parent-with-child sessions where the adult is driving the interaction and modeling what good AI use looks like. The mastery-learning platforms (Khan Academy Kids, Khanmigo, Beast Academy, IXL) are designed for direct child use and are appropriate. Open-ended conversational AI (ChatGPT, Claude) is a bigger risk for young children — content moderation isn’t bulletproof and the experience can pull kids out of the patient, slow, deeply attentive learning that’s developmentally appropriate.

Will AI replace teachers?

Not entirely, not soon. AI will replace some teaching tasks (drilling math facts, grading homework, providing first-pass writing feedback) reliably. It will not replace the work of knowing a specific child, holding them to standards, creating community in a classroom, and bringing moral authority to the relationship. Schools that fire their teachers and replace them with AI tutors will fail. Schools that keep their teachers, free them from grunt work via AI, and let them spend more time on what only humans can do will pull ahead.

My kid uses ChatGPT to do their homework. What should I do?

Talk to them. The right question is not “should you use AI?” — they will. The question is whether they’re using it to short-circuit learning or to accelerate it. Short-circuit: “Write my essay.” Accelerate: “Give me three different ways to think about this prompt before I write.” Sit with your kid for one homework session and watch what they’re actually doing. Adjust from there. Don’t ban it; teach it.

How much screen time is too much when AI is the learning tool?

AAP and other developmental-pediatrics groups have moved away from blanket hour-count limits toward content-type and context recommendations. A child doing 90 minutes of MathAcademy on a tablet is in a different category than 90 minutes of YouTube Shorts. That said, the physical-book and handwriting time still has to be non-zero. A rough heuristic for elementary kids: an hour-plus of focused AI-mediated learning is fine; another hour-plus of physical-book reading and pencil work the same day is necessary, not optional.

Is Alpha School worth the tuition?

For the right family in the right city — possibly. Alpha’s claimed results are real even after discounting for selection bias (children who attend Alpha come from families who care a lot about education). For most families, the better question is how much of the Alpha effect can be obtained at home with $200/month in tools and a present parent. The answer is “most of it,” and we have a full post on exactly how.

Where do physical books fit in an AI-heavy learning environment?

Center stage, not the supporting cast. Children who grow up reading physical books read more, retain more, and develop deeper attention. AI tools support the reading life; they do not substitute for it. The dedicated post walks through the research and what to do with it.

What if I’m an adult who never finished high school or college?

This is the audience AI changes the most. The combination of a free LLM, the Khan Academy GED prep track, and a couple of mastery-learning platforms is sufficient to credibly prepare an adult for the GED, for community-college placement tests, for the first two semesters of college math, and for most career-pivot skill tests. Time and discipline are the constraints. The information is no longer the constraint.

Sources

• Benjamin S. Bloom — “The 2 Sigma Problem” (1984, Educational Researcher)
• Pam A. Mueller & Daniel M. Oppenheimer — “The Pen Is Mightier Than the Keyboard” (2014, Psychological Science)
• Henry L. Roediger III & Mark A. McDaniel — Make It Stick: The Science of Successful Learning (Harvard University Press, 2014)
• John Sweller — Cognitive Load Theory (foundational work, ongoing publication record)
• Maryanne Wolf — Reader, Come Home: The Reading Brain in a Digital World (HarperCollins, 2018)
• Anne Mangen et al. — research on screen-vs-paper reading comprehension
• Naomi S. Baron — Words Onscreen: The Fate of Reading in a Digital World (Oxford University Press, 2015)
• Robert Bjork — UCLA “Bjork Learning and Forgetting Lab” — Desirable Difficulties research
• Dorothy L. Sayers — “The Lost Tools of Learning” (1947 Oxford address)
• National Center for Education Statistics — NAEP / Nation’s Report Card
• OECD — Programme for International Student Assessment (PISA)
• OECD — PIAAC adult literacy assessment
• Eric A. Hanushek & Harvard Center for Education Policy Research — Pandemic learning-loss research
• Emily Hanford / American Public Media — “Sold a Story” reporting on reading instruction
• Alpha School — 2 Hour Learning model and reported results
• Khan Academy — Khan Academy and Khanmigo
• MathAcademy.com — Mastery-learning math platform

You may also like

• Why Physical Books and Handwriting Still Matter for Learning
• Source of Truth: Physical Archives and Link Rot
• Alpha School Explained
• 2 Hour Learning Explained
• How to Approximate Alpha School at Home for Under $200/Month
• MathAcademy.com Explained
• AI for Homeschooling: The Complete Guide
• AI for Classical Homeschooling (with Dorothy Sayers)
• What Is a Large Language Model?