In 1984, University of Chicago psychologist Benjamin Bloom described a problem that looked almost unfair: students receiving one-to-one tutoring performed about two standard deviations better than conventional classroom students, which put the average tutored student above 98% of the control group. The barrier was never pedagogy. It was cost. Human tutoring at roughly $50 to $150 an hour does not scale nationally without turning into a multibillion-dollar expense.
That problem sat there for 40 years. The most credible answer so far is not a new school model. It is Khanmigo.
Sal Khan has said that when he saw GPT-4 in the summer of 2022, months before ChatGPT went mainstream, he could not sleep that weekend because the implication was obvious: software had crossed from delivering content to adapting in dialogue. Khan Academy moved fast, becoming OpenAI's launch partner for GPT-4 in March 2023 and building Khanmigo around a simple rule: do not hand over the answer. Ask questions. Push the student to explain. Keep going until the student can work it out.
A tutor, not an answer key
That design choice is the whole story.
Khanmigo uses Socratic questioning because an AI that simply outputs answers trains dependency, not understanding. Khan Academy has been explicit about that from the start: the product is meant to guide students through a problem rather than solve the problem for them. That is why the pricing matters. Families can get it for $4 per month, and grassroots teachers have free access in 70-plus countries.
The scale is no longer hypothetical. Khan Academy's SY24-25 annual report says Khanmigo grew 731% year over year to 2 million students, parents, and educators, while the broader platform reached roughly 120 million yearly learners and logged 7.7 billion learning minutes. K-12 Dive reported that student adoption in formal school settings climbed from 40,000 to 700,000 in a year. In January 2026, Khan Academy also expanded its AI stack through a Google partnership that put Gemini behind new Writing Coach and Reading Coach products for middle and high school students.
The research story is catching up to the product story. A January 2026 PNAS study from Stanford and the University of Toronto analyzed about 200,000 students and found that greater Khan Academy usage was associated with measurably higher math test scores. That does not prove AI tutoring solves every education problem. It does show that guided practice at scale is producing real learning outcomes, not just engagement metrics.
Chegg ran the opposite play
Chegg's business was built on a different premise: students will pay $15 to $20 a month for fast homework answers. That model worked until a general-purpose AI assistant made fast answers abundant and, in many cases, free.
Once ChatGPT shipped, Chegg was exposed. If the product is primarily a warehouse of pre-written solutions, a conversational model can replicate the perceived value instantly. The market has treated that as a fatal weakness. Chegg's stock is down about 99% from its 2021 peak, wiping out roughly $14.5 billion in value. The company has lost more than 500,000 subscribers since ChatGPT launched. In 2025 it cut staff twice, first by 22% in May and then by 45% in October, roughly half the workforce across the year. As of March 2026, services tracking analyst coverage still show a consensus rating of "Reduce".
Chegg did not fail because AI arrived. Chegg failed because its core value proposition was too close to what general-purpose AI gives away for free.
The market is voting for systems that improve people
That distinction is getting priced in across the category. Estimates now put the AI tutoring market at about $1.6 billion in 2024, growing to roughly $8 billion by 2030, while the broader AI-in-education market is projected to expand from about $5.9 billion to $32 billion over the same stretch. Those forecasts are noisy, but the direction is not. Buyers are moving toward products that personalize instruction, feedback, and coaching. They are moving away from products that just compress lookup time.
Khan Academy is unusually well positioned to capture that shift because the nonprofit model reinforces the product strategy. The organization brings in about $107 million a year and says 84% goes directly to educational programs. That keeps incentives aligned with learning outcomes instead of answer consumption. It also helps explain why Sal Khan's profile kept rising, including TED naming him its "vision steward" in October 2025.
The business lesson travels far beyond education
Most companies are making the same strategic choice now, whether they realize it or not. They can use AI to replace effort for the customer, or they can use AI to increase the customer's capability.
The first path looks attractive because it demos well. It feels magical when software spits out an answer, a draft, or a recommendation in three seconds. The problem is that answer engines are easy to commoditize. The moment a foundation model or a lower-cost competitor can do the same trick, your differentiation collapses.
Capability-building systems are harder to design. They require feedback loops, scaffolding, memory, and enough restraint not to do the job for the user. But they create stickier value because the customer gets better while using the product. That applies to tutoring, sales coaching, employee training, onboarding, analytics, even customer support. The winning AI product is often the one that makes the human sharper, not the one that makes the human passive.
Bloom's 2 Sigma Problem did not get solved by building a cheaper answer sheet. It got solved, or at least seriously attacked, by building a tutor that behaves like a tutor. Khan Academy understood that. Chegg did not. That is why one organization is compounding at scale with a $4 product and the other spent 2025 proving how quickly an answer business can unravel once AI makes answers cheap.