Skip to content
Published on

AI EdTech & Tutoring Tools 2026 - Khanmigo, MagicSchool, EduAide, MathGPT, Quizlet Q-Chat, Duolingo Max, Speak, Sizzle Deep Dive

🇰🇷KO🇺🇸EN🇯🇵JA
Split View필사 모드
Authors

Prologue — Why AI Tutoring Now

A classroom scene from November 2023. A high school student opens ChatGPT and types "solve this calculus problem." An answer comes out. Sometimes it's correct, sometimes not. The student copies it verbatim and submits it. The teacher grades it. Everyone knows something is going wrong. Learning is not happening.

The scene in May 2026 is different. The same student opens Khan Academy's Khanmigo and asks the same problem. Khanmigo doesn't give the answer. Instead, it asks back: "To solve this problem, let's first revisit the definition of a derivative. What does f'(x) mean?" When the student responds, the next question depends on whether their answer is right or wrong. This is Socratic tutoring, and it's the first serious pedagogical design decision AI has made.

This article is a May 2026 snapshot of that landscape. Which tools actually cause learning, and which are merely answer-vending machines? Price and feature comparisons will be obsolete in six months, but pedagogical design decisions and data privacy postures won't. We line up 30+ tools and gather the information students, teachers, administrators, and parents need to make decisions.

The key question for AI tutoring is not "can the AI give an answer" but "does the AI cause learning to happen." Those two often go in opposite directions.

Chapter 1 · Why AI Tutoring — Bloom's 2-Sigma Problem

In 1984, University of Chicago educational psychologist Benjamin Bloom published a famous paper titled "The 2 Sigma Problem." The conclusion was startling — students who received one-on-one mastery tutoring scored on average 2 standard deviations (98th percentile) higher than students in a conventional classroom. Two whole standard deviations. If the average classroom student scored 50 out of 100, the tutored group averaged 90+. This is an almost unprecedented effect size for an educational intervention.

The problem was scalability. Assigning one tutor to every student is not a model that can support mass education. Bloom's paper title — "The 2 Sigma Problem" — points exactly to this challenge: how can we replicate that 2-sigma effect inside a classroom? For 40 years, various attempts (peer tutoring, mastery learning, computer-assisted instruction) failed to come close.

The 2024-2026 hypothesis is clear. LLM-based AI can provide a scalable approximation of one-on-one tutoring. Finally, every student can have a "personal tutor" at an affordable price point. Khan Academy founder Sal Khan presented this vision at TED in 2023, and Khanmigo has become the most visible experiment of that bet from 2024 to 2026.

Evidence that the hypothesis is correct remains incomplete. Early studies report effects between 0.2 and 0.5 sigma — meaningful, but far from 2-sigma. Still, the direction is clear, and market bets are enormous. Khanmigo, MagicSchool, EduAide, Duolingo Max, Speak, Sizzle and dozens of others are testing the same hypothesis from different angles.

Chapter 2 · Khanmigo — The AI Tutor Khan Academy Defined

Khanmigo launched in March 2023 in beta from the Khan Academy / OpenAI partnership, and after general release in 2024 it has become the design reference point for K-12 AI tutoring as of 2026.

Design philosophy — Khanmigo's most decisive choice is the pedagogical principle of not giving answers. When a student asks "what's the answer?", Khanmigo refuses and asks back, "How might we start?" This is not GPT's default behavior. Khan Academy engineered it into the model through system prompts and reinforcement learning. It's the first major case of a pedagogical decision being codified as model behavior.

Feature areas — math (concept explanations, step-by-step coaching, wrong-answer breakdowns), writing tutor (essay feedback, argument structure review), debate coach (practice arguing both sides of a position), vocabulary builder, career coaching (job exploration with salary charts).

Pricing (May 2026) — Family donation model starting at 4 USD per month or 44 USD per year. School/district pricing around 35 USD per student per year (pilot basis). The pricing reflects Khan Academy's nonprofit identity, decisively cheaper than most competitors.

LAUSD pilot — In 2024 Los Angeles Unified School District (LAUSD) signed a partnership with Khan Academy and rolled out Khanmigo to roughly 100,000 students as a pilot. Some result data was released in spring 2026, reporting statistically significant gains in math achievement (exact effect sizes vary by grade and subject).

Limitations and honest evaluation — Khanmigo is tightly coupled to Khan Academy content. It is weak at analyzing arbitrary external materials. English is its primary language, and response quality drops in non-English contexts like Korean or Japanese. The biggest limitation is math hallucination — in complex calculus or linear algebra, steps occasionally go wrong. Just often enough that a student can sometimes catch it.

Chapter 3 · MagicSchool — A Suite of 60+ Tools for Teachers

MagicSchool (magicschool.ai) bets not on students but on teacher productivity. It automates the repetitive daily tasks teachers perform: grading, lesson planning, parent communication, IEP (Individualized Education Plan) drafts, and more.

Tool catalog — As of May 2026, the catalog lists more than 60 micro tools. Lesson plan generator, unit assessment generator, student IEP draft writer, parent email tone adjuster, text leveling tool (rewriting the same material at grade 3, 6, or 9 level), discussion question generator, rubric builder, grading assistant, and more. Each tool is single-purpose, designed so a teacher can use it in five minutes and move on.

Pricing (May 2026) — Free plan (30 uses per month), Plus 14.99 USD per month (for individual teachers), Schools/Districts negotiated (often around 79 USD per teacher per year). The free plan is unexpectedly generous, making it easy for small schools or individual teachers to start.

Pedagogical perspective — Because MagicSchool helps teachers rather than students, it side-steps the criticism that "AI is stealing learning." That said, outputs like IEP drafts or parent emails ultimately affect students' lives. The key is the workflow where teachers review and take responsibility for AI output, not pass it through verbatim. MagicSchool itself emphasizes this — "AI drafts, teachers decide."

Honest evaluation — The most mature tool in the category. Time savings are obvious, and the ROI on a single license is immediately visible. That said, 60 tools means 60 things to learn, and in practice each teacher converges to 5-7 favorites.

Chapter 4 · EduAide — Another Face of Teacher Assistant AI

EduAide (eduaide.ai) is in the same category as MagicSchool (teacher productivity) but takes a different approach. It focuses on lesson planning workflows.

EduAide's core is the "Teaching Assistant" chat interface. When a teacher types "lesson plan for introducing linear functions in 8th grade math," EduAide generates learning objectives, pre-assessment, core activities, differentiation strategies, and formative assessment in one pass. Within the same context, it can then produce unit assessments, student worksheets, and classroom activities.

Pricing — Free plan (25 content generations per month), Pro 12 USD per month (unlimited), Schools/Districts negotiated. Comparable to MagicSchool, but EduAide invests more in workflow depth within a single workflow.

How to choose — MagicSchool is breadth (60 tools); EduAide is depth (unit-level workflow). Teachers who try both say "it depends on which one matches your mental model." Both have free tiers, so a one-semester trial is the fastest way to decide.

Chapter 5 · Quizlet Q-Chat — The Flashcard's AI Friend

Quizlet has been synonymous with flashcard learning since 2005, and it joined the AI era in 2023 with Q-Chat. Q-Chat is a GPT-based learning partner that operates on top of user-created card sets.

How it works — When a user creates a biology card set, Q-Chat asks questions grounded in that set. "Name the two stages of photosynthesis," and when the user answers it provides immediate feedback. Not just rote testing, but also application questions — "If photosynthesis stopped, what would happen in a plant cell?"

Pricing — Quizlet free (with ads), Plus 7.99 USD per month or 35.99 USD per year (unlimited Q-Chat, ad removal, AI study guides). Reasonable in a student price range.

Limitations — Q-Chat strongly depends on the quality of user-created cards. If the cards are inaccurate, Q-Chat teaches inaccurately. And for foreign-language learning, pronunciation feedback is weak — purpose-built tools like ELSA Speak or Speak are much better there.

Chapter 6 · Duolingo Max — Grafting GPT-4 onto Gamification

Duolingo launched Duolingo Max in March 2023 in partnership with OpenAI, and as of 2026 it represents the largest-scale integration of GPT-4 into learning.

Two flagship features — (1) Explain My Answer — when the learner submits a wrong answer, GPT explains why it was wrong and shows the correct pattern. Not just a red "X" but coaching like "This verb conjugates in the past tense; you wrote it in the infinitive." (2) Roleplay — GPT plays roles like cafe server, airport agent, or friend, and converses with the learner. Learner responses are open-ended, and GPT evaluates appropriateness, grammar, and naturalness.

Pricing (May 2026) — Super Duolingo 14 USD per month or 84 USD per year (annual is efficient); Duolingo Max 19 USD per month or 168 USD per year. Max stacks GPT features on top of Super.

Limitations and pedagogical evaluation — Duolingo's essential limitation remains. Gamification creates motivation but doesn't translate to deep language ability. It's common for a five-year-daily Duolingo user to be unable to read a newspaper in that language. Adding GPT-4 only partially solves this. Serious learners typically start with Duolingo and then move to other tools (Pimsleur, Lingoda, iTalki) after 6-9 months.

Chapter 7 · Speak — The Y Combinator AI Conversation Champion

Speak (speak.com) is a Korean-American startup out of Y Combinator that bet from day one on voice conversation with AI. It received funding from OpenAI twice and is one of the few EdTech companies OpenAI has invested in directly.

Core design — Speak treats voice, not text, as the primary interface. The learner speaks, and the AI listens and replies. It analyzes the learner's pronunciation, identifies weak phonemes, and repeatedly exposes them to words containing those phonemes. The strongest feature is free-form speaking practice — not pre-scripted sentences, but the learner attempting whatever they want to say while the AI evaluates.

Korean success — Interestingly, even though Speak is US-based, a large share of users are in Korea. The Korean English-learning market — the pain point of "I can read it but I can't speak it" — fits the tool precisely. In Korea, a usage pattern of 20,000-30,000 KRW per month replacing English conversation academies has taken hold.

Pricing (May 2026) — Premium around 20 USD per month, roughly 99 USD per year. Korea has separate pricing that's sometimes cheaper.

Limitations — Speak's weakness is content variety. Business English, academic English, medical English, and other specialized domains are weak. It's general conversation first.

Chapter 8 · Praktika.ai — Conversation with Virtual Avatars

Praktika.ai is in the same category as Speak (AI conversation) but makes a different design decision: virtual avatars. The AI tutor has a face, expressions, and a moving mouth.

The value of this decision is debated. Some users feel they speak more naturally because there's an avatar; others find "designing me to treat a thing I know is an AI like a human feels awkward." Pronunciation and grammar feedback is comparable to Speak.

Pricing — Subscription model around 14.99 USD per month.

Chapter 9 · Sizzle — AI Math Tutor

Sizzle (sizzle.ai) is a math-only AI tutor built by ex-Google engineers. If Khanmigo is K-12 general, Sizzle focuses on the single domain of math.

Design decision — When a student photographs a problem, Sizzle coaches the solution step by step. It chose the no-answer pedagogy — give a first hint, and when the student completes that step give the next hint. Similar to Khanmigo's Socratic style.

Audience — From 6th grade through college math (calculus, linear algebra) in US grade terms. Pricing is free (ad-supported) or Pro around 5 USD per month or 30 USD per year. Reasonable for a single-domain tool.

Limitations — English is the primary language. Math problems entered in Korean lose accuracy. And photo input is far more reliable than typed input — OCR quality is more stable than parsing text-form math expressions.

Chapter 10 · Photomath — The Math OCR Standard Google Acquired

Photomath, founded in 2014 as a Croatian startup, has been the design reference for the "take a photo, solve the math" app. Google acquired it for approximately 100 million USD in 2022, and as of 2026 it is being integrated into Google Search, Lens, and education tools.

How it works — A student photographs a math problem; OCR recognizes the expression and step-by-step solutions are shown. Unlike Khanmigo, Photomath shows answers immediately. Photomath's design decision prioritizes convenience over pedagogy.

Pricing — Free use available; Plus 9.99 USD per month or 69.99 USD per year (step-by-step explanation videos, ad removal).

Pedagogical criticism — Photomath is the tool most frequently criticized as an "AI cheating tool." There are many cases of students copying homework answers verbatim. Post-Google acquisition there have been promises to strengthen the learning mode (step-by-step coaching), and improvements have indeed shipped, but the fundamental criticism remains.

Chapter 11 · Mathpix — From Image to LaTeX

Mathpix (mathpix.com) is not a learning tool but infrastructure. It converts math-expression images to LaTeX code. Graduate students, researchers, and teachers use it when porting equations from papers or textbooks.

Pricing — Free (50 snapshots per month), personal Pro starting around 4.99 USD per month (1,000 snapshots), API usage billed separately. Reasonable for students and researchers.

Why it's in the EdTech category — Mathpix itself is not tutoring, but it serves as a bridge to deliver equations accurately to other AI tools (GPT, Claude). The graduate-student workflow "take a photo, extract LaTeX with Mathpix, paste into GPT/Claude for explanation" has become standard.

Chapter 12 · Wolfram Alpha — The 30-Year-Old Symbolic Math Engine

Wolfram Alpha, launched in 2009 as a "computational knowledge engine," was a tool for getting mathematical answers to natural-language queries long before the LLM era. After ChatGPT appeared it seemed briefly threatened, but as of 2026 it has settled into a complementary role to LLMs.

Why LLMs cannot replace Wolfram — LLMs reason about math probabilistically. They hallucinate. For complex integrals or differential equations, LLM answers are often wrong. Wolfram Alpha is a symbolic computation engine; it executes defined algorithms. Its answers are always correct.

Wolfram + ChatGPT integration — Since 2023, ChatGPT plugins / tool calls frequently invoke Wolfram. When an LLM receives a math problem it can delegate the call to Wolfram and explain Wolfram's answer in natural language. This architecture dramatically reduces hallucination.

Pricing — Wolfram Alpha free (basic results), Pro 7.25 USD per month (step-by-step solutions plus more compute), Pro Premium 9.99 USD per month (additional API access). Pro is enough for students.

Chapter 13 · Coursera Coach and LinkedIn Learning Coach — Corporate and Lifelong Learning

Beyond K-12, AI is also taking root in adult lifelong learning and corporate training. Coursera Coach and LinkedIn Learning Coach are the prime examples.

Coursera Coach — When a learner asks "I don't understand this part" while watching a lecture video, or asks the whole course "help me prepare for the final exam," Coach grounds its answer in the video transcript and course context. The difference from generic ChatGPT is course context grounding.

LinkedIn Learning Coach — Same idea applied to LinkedIn courses. Additionally it reads the user's LinkedIn profile and evaluates "is this course appropriate given your job role."

Pricing — Coursera Plus 59 USD per month or 399 USD per year (Coach included). LinkedIn Learning 39.99 USD per month (Coach included). Corporate licenses negotiated separately.

Limitations — Both tools work best for English courses. Course-context grounding is weaker for Korean or Japanese lecture content.

Chapter 14 · edX with ChatGPT and Udemy AI — Other MOOCs

edX — In 2024 edX launched a ChatGPT-based learning assistant, and as of 2026 it is integrated across all courses. Operating atop MIT and Harvard content is the advantage.

Udemy — Udemy launched a similar AI assistant around the same time. The quality variance across Udemy courses is so wide, however, that even with AI grounding the result-quality variance persists.

The decision is simple: which platform's courses are you watching? That dictates which tool follows along.

Chapter 15 · MathGPT — Math-Specialized LLM

MathGPT is the umbrella term for tools billed as "LLMs specialized for math." It's not a single company but several variants. The most well-known is the commercial product at mathgptpro.com, which claims strong step-by-step math reasoning.

Hallucination problem — Marketed as more accurate at math than general GPT, but independent evaluations show small differences. Serious math study is still best served by the combination of Wolfram + Khanmigo + a human tutor.

Pricing — Around 14.99 USD per month. Not a particularly compelling price point.

Chapter 16 · Chegg, Course Hero, Brainly, GauthMath — The Answer-Engine Camp

This category is pedagogically the most contested. Their business models are essentially "let students get homework answers fast," and adding AI didn't change that essence.

Chegg — User base over 100 million. After ChatGPT launched in late 2023, revenue wobbled significantly, and Chegg responded with its own AI (CheggMate, later Chegg AI Companion). Pricing around 19.95 USD per month. The most widely used among students, and also the most frequently mentioned in academic integrity cases.

Course Hero — Platform for sharing college lecture notes and assignment answers; added an AI assistant. Pricing 39.95 USD per month.

Brainly — Community Q&A + AI solver. Added AI answers on top of the peer-help model. Free base, Plus around 4 USD per month.

GauthMath — Photo input + AI math solver. A Photomath competitor. Free + Plus tier.

Honest evaluation — These tools are hard to defend as "good for learning." Rather than learning effects, they contribute to time savings (charitably) or academic dishonesty (strictly). The point isn't "students shouldn't use them," but rather "use them with eyes open about what they are."

Chapter 17 · Knewton, Smart Sparrow, DreamBox, IXL, Prodigy — The Old Adaptive-Learning Camp

EdTech existed before LLMs. These are tools that built the market with adaptive learning in the 1990s-2010s, and the 2026 question is how they adapt to the LLM era.

  • Knewton — Acquired by Wiley, now integrated into the alta platform. College-level adaptive learning.
  • Smart Sparrow — Acquired by Pearson, integrated as a learning design tool.
  • DreamBox — K-8 math and reading adaptive platform. District licensing. Around 50-80 USD per student per year.
  • IXL — K-12 all-subject practice platform. Family subscriptions around 19.95 USD per month; district pricing separate.
  • Prodigy — K-8 math game in RPG form. Free base, Premium around 9-15 USD per month.

Adapting to LLMs — Their strength is content catalogs and learning data built over decades. Layering LLMs on top yields powerful tools. But their integration pace lags new LLM-native tools, and market share is gradually slipping.

Chapter 18 · Synthesis Tutor, CK-12, Replit, Codecademy, Brilliant — Other Camps

Synthesis Tutor — From Synthesis School, originally built for Elon Musk's children. Specialized in age 5-11 math. Pricing in the expensive range of 29-49 USD per month, but with deep pedagogical design.

CK-12 — Open-textbook integration with AI. Free is the core value proposition. K-12 all subjects.

Replit Teams + AI — AI is a huge variable in coding education. Replit provides classroom licenses giving students IDE + AI coding assistant.

Codecademy + Cody AI — Codecademy added its own AI (Cody). Interactive coding courses with step-by-step coaching.

Brilliant.org — Interactive learning platform for math, science, and CS. Visualization is the strength. AI integration is comparatively light. Around 149 USD per year.

Chapter 19 · English Learning — Cambly, Preply, iTalki, Lingoda, Babbel, Rosetta Stone, ELSA Speak

English (or foreign-language) learning is one of EdTech's biggest markets, and in 2026 the hybrid of human tutor + AI assistant is becoming the norm.

  • Cambly — 1-on-1 video calls with native English speakers. Per-minute pricing. Monthly subscription around 30-150 USD.
  • Preply — Marketplace of human tutors across many languages. Hourly rates from 5 to 100 USD.
  • iTalki — Preply competitor. Similar model.
  • Lingoda — Group classes + 1-on-1. Around 50-200 USD per month.
  • Babbel — App-based language learning. Around 14 USD per month. Live class option separate.
  • Rosetta Stone — The oldest language-learning brand. Around 11.99 USD per month.
  • ELSA Speak — AI for English pronunciation correction. Phoneme-level analysis. Pro around 11.99 USD per month. Large user bases in Korea and Vietnam.

Strategy guide — Lay the foundation with Duolingo, sharpen pronunciation with ELSA Speak, do conversation with human tutors on iTalki / Preply, get real exposure to varied native speakers on Cambly, finish with Lingoda group classes. This sequence is the most efficient English-learning path in 2026.

Chapter 20 · STEM Simulation — Wolfram Mathematica, GeoGebra, Desmos, PhET

Simulation tools form a separate category in STEM education.

  • Wolfram Mathematica — Integrated symbolic, numeric, and visualization. University / research level. Student license around 69 USD per year.
  • GeoGebra — Free dynamic geometry, algebra, and calculus visualization. Widely used in schools worldwide.
  • Desmos — Free graphing calculator. Added classroom activity builder (Polygraph etc.).
  • PhET (University of Colorado) — Free interactive physics, chemistry, biology simulations. Catalog of over 1,000 simulations.

AI integration — These tools have integrated AI relatively slowly. Wolfram leads on LLM integration; GeoGebra, Desmos, and PhET continue to invest more in the value of their content itself.

Chapter 21 · Korean AI EdTech — QANDA, Mathking, Mathflat, Classting, Tongtong

Korea has a massive private-education / EdTech market and moves fast on AI integration.

  • MathPresso QANDA — Korea's flagship math AI service. Photo input - AI solution. Expanding globally. Free + Plus pricing. Large user base in Korea, Japan, and Southeast Asia.
  • Mathking, Mathflat — Tools specialized for Korean college-entrance math, with AI solving plus learning analytics.
  • Classting — LMS platform for K-12 schools; adding AI features.
  • Tongtong — A note-taking helper AI; recognizes student handwriting and provides learning analytics.

Pedagogical evaluation — Korean tools operate in a market where "getting the right answer for the exam" is the primary goal, so they focus on accuracy and speed over pedagogical depth. This reflects Korean study culture, and is also the limit of the Korean market.

Chapter 22 · Japanese EdTech — Z-kai, Benesse, Studyplus, Manabo, Mirai Coach

Japan has a different EdTech culture from Korea. The long tradition of correspondence education (mail / online) is migrating to digital.

  • Z-kai — Founded in 1931, a prestigious correspondence-education name. Expanding into digital + AI tutoring.
  • Benesse — Along with Z-kai, one of the two giants of Japanese correspondence education. Adding AI assistants to the "Shinken Zemi" series.
  • Studyplus — Study-time tracking + study-habit management. Added AI-based learning analytics.
  • Manabo — 24-hour AI home-tutor concept.
  • Mirai Coach — Career counseling integrated with AI.

The Japanese market's characteristic is trust-based gradual adoption. New tools spread slowly; existing incumbents (Z-kai, Benesse) absorb AI.

Chapter 23 · University Adoption — ASU + OpenAI, Stanford Co-Teacher

Separate from K-12, universities are also adopting AI EdTech seriously.

ASU + OpenAI — In January 2024 Arizona State University (ASU) officially announced ChatGPT Enterprise adoption with OpenAI. Across 75,000 undergraduates, the first large-scale rollout of ChatGPT-based learning tools at the institutional level.

Stanford Co-Teacher — Stanford internally developed a GPT-based teaching-assistant tool, currently in pilot at some courses.

MIT, Harvard, and others — Most build their own policies and decide on use / non-use per course. A model of per-course choice rather than uniform institutional rollout.

Chapter 24 · Concern 1 — Cheating and Turnitin AI Detector

The most visible concern with AI EdTech is academic dishonesty. The scenario where a student has the AI write an essay and submits it under their own name.

Turnitin AI Writing Detector — Launched in 2023. Provides a score representing AI-generation probability. But accuracy debates continue. Both false positives (flagging human writing as AI) and false negatives (missing AI writing) are reported. Criticism is especially loud that ESL (non-native English) students are vulnerable to false positives.

Practical guidance — Don't use a Turnitin AI score as standalone evidence of academic dishonesty. It must be considered alongside other signals (the student's prior writing, oral responses). Schools should establish explicit policies and clearly tell students "what AI use is permitted."

Chapter 25 · Concern 2 — Hallucination and Student Data Privacy

Hallucinations — LLMs frequently err in math and science. If a student lacks the ability to verify the AI's answer, hallucination directly damages learning. This is why Khanmigo and Sizzle chose the "no-answer" design.

Student data privacy — In the US, FERPA (Family Educational Rights and Privacy Act); in Europe, GDPR; for minors, COPPA (Children's Online Privacy Protection Act, under 13). Schools must comply with these regulations when adopting AI tools. Yet many AI tools' terms explicitly say "we may use user input for model training." Applied to minor data, that creates COPPA risk.

Practical guidance — For school / district adoption, explicitly verify (a) opt-out of training, (b) data retention periods, (c) third-party sharing policies. Individual teachers using ChatGPT free tier with student data is an obvious risk.

Chapter 26 · AI Grading — PEG Writing, Gradescope, EssayCheck

AI has entered not just learning but also grading.

  • PEG Writing — Automated essay scoring system originating in the 1960s, retooled for the LLM era.
  • Gradescope — Workflow tool for grading college exams. Partially integrates AI-based auto-grading.
  • EssayCheck — Automated essay feedback.

Concerns — AI grading offers value in (a) grading consistency, (b) time savings. But (c) pedagogical feedback quality, and (d) grading bias (potential discrimination by race, language, gender) make it unable to replace human teachers. The current consensus is a hybrid model: "AI first pass, human final decision."

Chapter 27 · Price Comparison — One-Page Table

Major tools' monthly pricing as of May 2026:

ToolMonthly PriceCategory
Khanmigo4 USDK-12 AI tutor
MagicSchool Plus14.99 USDTeacher productivity
EduAide Pro12 USDTeacher productivity
Quizlet Plus7.99 USDFlashcards + AI
Duolingo Max19 USDLanguage learning
Speak Premium20 USDEnglish conversation AI
Sizzle Pro5 USDMath AI
Photomath Plus9.99 USDMath OCR
Mathpix Pro4.99 USDEquation OCR
Wolfram Alpha Pro7.25 USDSymbolic math
Coursera Plus59 USDMOOC
LinkedIn Learning39.99 USDCorporate learning
Chegg Study19.95 USDHomework answers
Course Hero Premier39.95 USDHomework answers
Brainly Plus4 USDQ&A + AI
IXL Family19.95 USDK-12 practice
Brilliant12 USDMath/Science/CS
Cambly30-150 USDEnglish conversation
PreplyPer tutorHuman tutor
ELSA Speak Pro11.99 USDEnglish pronunciation
Babbel14 USDLanguage
Rosetta Stone11.99 USDLanguage
Wolfram Mathematica Student5.75 USDSymbolic math

Selection guide — You don't need to buy everything for one student. Pick one or two learning goals and use 1-3 tools deeply for them. That has more effect.

Chapter 28 · Conclusion — Has the 2-Sigma Problem Been Solved?

In spring 2026, the answer to Bloom's 2-sigma hypothesis is not yet clear. Measurable effects (0.2-0.5 sigma) are evident. The full 2-sigma effect (0.5-2.0) is unproven. But the direction is correct.

For students, the recommendations are simple — (1) pick tools that cause learning, not tools that give answers. Khanmigo, Sizzle, Speak fall on this side. The answer-engine modes of Chegg and Photomath damage learning. (2) Use one tool deeply. Using one tool deeply for 90 days yields more effect than using five tools shallowly. (3) Don't forget humans. The value of teachers, peers, and human tutors is not replaced by AI.

For teachers, the recommendations are also simple — (1) buy time with teacher-productivity tools like MagicSchool and EduAide. Spend that time on personalized student feedback. (2) Don't pass AI output to students unedited. Always review and take responsibility. (3) Make AI-use rules explicit to students. Ambiguity quickly becomes academic-integrity trouble.

For school administrators — verify FERPA/GDPR/COPPA compliance, negotiate district licenses, invest in teacher training. The pattern of individual teachers ad-hoc adopting free tools carries large risk.

2-sigma is not yet solved. But we live in the first generation that can intentionally design toward 2-sigma. Those design decisions — pedagogy, privacy, pricing — will shape education for the next decade.

Chapter 29 · References

Discuss on Twitter • View on GitHub