Surviving AI in Education: From Answer Provider to Learning Designer

Editorial Note: This essay examines the structural contraction of the “answer economy” and the concurrent repricing of human-led education in the age of generative AI. The opening scenes and industry profiles are composite narratives synthesized from documented market conditions and publicly reported trends; they are not presented as accounts of identifiable individuals. Market data, corporate disclosures, and research findings cited in the piece are drawn from public records and published research. The piece is intended as an analytical framework for educators, policymakers, and workers navigating this transition, including a set of forward-looking scenarios rather than a point forecast.

I. The “Answer Economy”—the market for standardized, on-demand educational answers

In the autumn of 2022, inside a drafty shared apartment off the University of Michigan campus, a twenty-seven-year-old materials-science Ph.D. candidate settled into an ironclad routine. Three nights a week, he logged into the backend portal of Chegg Study, claimed a queue of pending physics and calculus problems, and drafted step-by-step solutions. He submitted them for review, and for every answer that passed quality control, he was paid.

It was a quiet, reliable hustle. The gig brought in between $1,200 and $1,500 a month. For a graduate student living on a stipend, that margin was the difference between anxiety and stability; it covered rent and health insurance with room to spare. The work was predictable. The queue was perpetually full, the grading rubrics were consistent, and the demand seemed infinite. American undergraduates, it appeared, would never stop needing someone to show them the work.

The shift began in the spring of 2023. The first anomaly was not a change in policy, but a sudden, inexplicable thinning of the queue. His acceptance rate remained flawless. The platform’s grading standards hadn’t changed. The competition hadn’t stiffened. Yet, week by week, the available problems simply evaporated. He received no email explaining the drop. The platform announced no new rules. The demand had not migrated; it had vanished into thin air.

By 2025, the macroeconomic picture of the company he worked for had become a cautionary tale. Chegg went through two major workforce reductions in 2025, while its business contracted sharply as students increasingly turned to generative AI and search engines for academic support. The company did not disappear from homework help overnight; instead, it substantially reduced the scale of its traditional Academic Services business while shifting strategic emphasis toward professional skilling. The salaried employees who lost their jobs received severance packages, dignified farewell emails, and sympathetic coverage in the business press. But the contract experts—the workers whose income depended on the flow of questions through the platform—were not part of the company's employee headcount. The word "layoff" implies an employment relationship that never existed. For them, the gig could simply become less viable as the queue contracted.

This article begins with that phantom worker. But the goal is not to rehash the exhausted, binary debate over whether artificial intelligence will "replace teachers." That question has been argued to the point of meaninglessness. The reality is far more granular and far more urgent. What actually demands our attention are three specific things: why the specific business model this student represented collapsed, exactly where the boundary of that collapse was drawn, and where the people still trying to make a living in this industry should place their chips over the next twenty-four months.

II. The First Domino

Chegg’s collapse is deeply instructive because it was public, dramatic, and early.

During the earnings call in May 2023, Chief Executive Dan Rosensweig made a striking admission: since March, student interest had visibly shifted toward ChatGPT, which was "negatively impacting our new customer growth rate." Chegg became one of the earliest major publicly traded companies to explicitly attribute a financial setback to generative AI on an earnings call. The market’s response was swift; the stock fell by roughly 47 percent in early trading.

What followed was a prolonged corporate contraction. Chegg’s stock eventually fell more than 99 percent from its 2021 peak, while the company’s subscription base and revenue declined sharply. Subscription Services subscribers fell from 7.7 million in 2023 to 6.6 million in 2024, with the decline continuing into 2025. By the end of 2025, Chegg had reduced its employee headcount to 595, from 1,271 at the end of 2024.

But the broader labor-market evidence is also instructive. A peer-reviewed study of freelancers on a large online labor platform found that, after ChatGPT’s release, freelancers in writing-related occupations experienced a 2 percent decline in monthly jobs and a 5.2 percent decline in monthly earnings relative to less-exposed occupations. The study captures a specific platform and a short-term labor-market effect; it does not measure the education sector directly.

This is how the "Answer Economy" dies: not with a spectacular crash, not with a bankruptcy filing, but with a slow, unceremonious evaporation.

The pressure was not confined to Chegg. Other education and learning companies also went through layoffs, restructuring, or changes in investor expectations during the post-pandemic reset. But those cases should not be treated as direct evidence of generative-AI substitution: the causes varied, including changes in consumer demand, cost structures, and the broader funding environment. The more useful comparison is at the level of the product itself.

The businesses most exposed to AI substitution tend to share a different set of characteristics: their deliverables are highly standardized, time-sensitive, and relatively easy to evaluate by output alone. A step-by-step math solution, a slide-deck template, or a basic recorded explanation can often be judged without much knowledge of the person who produced it. That makes these products unusually vulnerable when a machine can generate a plausible substitute at very low marginal cost.

This leads to a blunt, unavoidable conclusion: what collapsed was not "education." What collapsed was the business of delivering answers on demand, in the correct format.

The gap between "education" and "an answer"—the messy, human work of diagnosis, motivation, companionship, and trust—used to be a blind spot that the answer-delivery business covered by default. Now that the answer is free, that blind spot has become the only territory the industry has left. The question, then, is what AI has actually done to that remaining territory.

III. The Illusion of the AI Classroom

To understand the current landscape, one must first confront the speed of AI adoption. By May 2025, 84 percent of U.S. high school students surveyed by the College Board reported using generative AI tools for schoolwork. MagicSchool, an AI-powered educator platform, reported 3 million educators actively using the platform, while Khan Academy reported that 2 million people worldwide—including students, educators, and parents—used Khanmigo during the 2024–25 school year.

Yet adoption does not automatically translate into learning gains. A 2026 two-year randomized study across 18 Tennessee middle schools found that students assigned access to Khanmigo achieved modest improvements in mathematics, while actual use of the AI tutor was relatively limited. Ninety-six percent of students tried Khanmigo at least once, but the median student messaged it on only about one-third of practice days and in only 17 percent of sessions in which they made a mistake.

It is difficult to overstate the significance of this admission. It came from the person with the least incentive to say it. Khan Academy had bet its future on the AI tutor; if Khanmigo were reliably moving the needle for the average student, no one would have been more eager to broadcast that victory than its founder.

When you hold the adoption data and the efficacy data up to the light simultaneously, the true silhouette of AI in the classroom comes into focus. AI’s actual role is not that of a replacement teacher; it is a leverage tool for the human teacher. It has driven the marginal cost of lesson planning, test generation, grading, and differentiated material creation down to near zero. The six million teachers using MagicSchool prove that educators are eagerly collecting this leverage.

But AI has not lowered the cost of getting a fifteen-year-old to sit down and actually learn.

The cost structure of diagnosing exactly why a student is stuck, maintaining their motivation through a frustrating Tuesday afternoon, managing classroom dynamics, and convincing parents that the process is working—these costs have not moved a single inch. Why? Because the price of these services is not determined by productive efficiency. It is determined by trust. A student can ask an AI a question at any hour of the day, but most of the time, they simply won't. And when they do ask, they rarely stick with the answer long enough to internalize it. The Khanmigo data perfectly illustrates this paradox: the tool is sitting right there, but the human mechanism required to activate it is missing.

To put it in economic terms: AI has made the transmission of knowledge free. But learning was never just a knowledge-transfer problem; it is a behavioral one. The space between those two sentences is where all the repricing in the education industry is currently happening.

IV. The Great Repricing of Human Instruction

Before proceeding, a necessary caveat: the protagonist of this next section is a composite. While every data point and market dynamic attached to her is backed by public market research, "she" is an archetype synthesized from industry trends. Anyone who spends time interviewing hundreds of independent educators will bump into this exact profile over and over again.

She lives in Austin, Texas, and provides one-on-one tutoring for the SAT and AP Chemistry. Around 2023, she noticed two seemingly contradictory things happening to her business at the exact same time: her waitlist was getting longer, and her hourly rate was climbing.

When you unpack the mechanics of her market, both phenomena make perfect, structural sense.

The first mechanism is the collapse of her cheapest competition. A family on a tight budget used to face a binary choice: pay $40 an hour for a low-tier tutor, or let the kid struggle alone. Now, the middle option is a free chatbot. The market for cheap, low-level explanation has been entirely hollowed out. But here is the catch: that demand did not stay with the AI. SAT and AP scores are directly tethered to college admissions outcomes. Families are not looking for a chatbot; they are looking for accountability, structure, and a human throat to choke if the scores don't improve. Consequently, the demand flowed back to human instructors—specifically, to independent tutors who could deliver an "institutional-grade" experience. When Chegg shut down its live tutoring arm in 2021, a wave of tutors spilled onto platforms like Wyzant to build solo practices. The people who were "liberated" from platform dependency back then are now standing at the very front of the pricing surge.

The second mechanism is AI acting as a force multiplier. AI allows this single tutor to deliver the kind of backend infrastructure that only a massive tutoring agency could previously afford. Before a session, she uses AI to generate a comprehensive diagnostic report for the student: pinpointing knowledge gaps, identifying recurring error patterns, and mapping out an eight-week customized learning trajectory. By the end of the very first session, the parents receive a polished, data-rich document that screams professionalism. After the lesson, AI instantly processes the student's mistakes into a personalized review sheet, which she emails to both the student and the parents. Her prep time can be cut substantially. She reinvests those saved hours directly into the two things AI cannot do: having deep, empathetic one-on-one conversations with the student, and providing high-frequency, reassuring feedback to the parents.

Market data confirms she is not an outlier. A September 2025 analysis by Wiingy, covering 3,600 platform tutors, found that the average rate for SAT tutoring was $62 an hour, but in highly competitive urban markets, rates easily exceeded $300 an hour, driven by the intensifying arms race of college admissions. On Wyzant, top-tier SAT tutors are listing their rates between $220 and $495 an hour, and their booking volumes are staggering; some have logged over 3,000 billable hours.

In this specific market, the price of standardized content is in freefall, while the price of bespoke human service is hitting record highs. Both things are happening simultaneously, and they are not contradictory.

This allows us to draw a new map of the industry. The market has fractured into three distinct tiers:

The Bottom: Free, generalized AI consumes all standard explanations.
The Top: Premium human instructors capture more value from judgment, strategy, accountability, and emotional connection.
The Middle: Standardized, paid content products (the "Answer Economy") are being hollowed out.

This three-tier model is a highly effective lens. If you hold it up to every sub-sector of the education industry, it becomes immediately obvious who is sitting at which table.

V. The Industry Map: Who Is Taking the Hit, and Why?

When you run the education sector through this three-tier lens, the damage assessment is remarkably clean.

Homework Help and Search Platforms (Impact: Total Collapse) This is the category with some of the clearest evidence of AI-driven pressure. Students increasingly use generative AI for homework and academic support, and Chegg has explicitly identified that shift as a major factor in declining traffic, subscriptions, and revenue. But the market has not disappeared entirely: Chegg still operates Academic Services, even as that business has contracted sharply and the company shifts its strategic emphasis toward skilling.

Language Learning - Instruction and Conversation (Impact: Severe) Consider Duolingo’s response to the changing AI landscape. In 2025, the company generated $1.04 billion in revenue, up 39 percent year over year. In 2026, it deliberately shifted some advanced AI-powered learning features, including Video Call with Lily, to a broader paid tier while also investing more heavily in free AI-powered speaking experiences. Management estimated that reducing monetization friction would mean more than $50 million in foregone bookings in 2026. This does not prove that AI is destroying the language-learning market, but it does show how AI is changing the economics of acquiring and monetizing learners.

Standardized Course Recording / General Ed Instructors (Impact: Severe) The evidence here is more mixed than the Chegg case. Coursera’s consumer revenue grew 10 percent in 2025, while Udemy’s consumer revenue declined 9 percent largely as the company shifted from transactional course purchases toward recurring subscriptions; its consumer subscription revenue grew 44 percent. At the same time, both platforms are increasingly incorporating AI into learning and content delivery. The more defensible conclusion is not that recorded courses are disappearing, but that static content is becoming less differentiated as AI makes explanations and practice increasingly interactive and personalized. The value of instructors may therefore shift toward curation, credibility, community, and applied guidance rather than disappearing altogether.

Instructional Design (Impact: Severe) This is a likely area of job compression. AI can now generate course outlines, quizzes, slide drafts, examples, and other instructional assets in seconds. The human contribution is increasingly concentrated in deciding which learning objectives, sequence, assessment strategy, and instructional approach fit a particular learner or context. As asset production becomes cheaper, the relative value of instructional judgment may rise even as fewer people are needed for routine production work.

Essay Coaching and Writing Center Consultants (Impact: Moderate to Severe) Proofreading, structural suggestions, and literature reviews—the core menu of writing center services—are directly in AI’s crosshairs. What survives is the strategic guidance that operates safely within the boundaries of academic integrity, and the oral tradition of teaching a student how to actually think about research.

Test Prep: The "Question-Walking" Segment (Impact: Moderate) The demand for mid-tier tutors to simply walk a student through practice problems has vanished. However, high-level strategic coaching—time management, test-day decision-making, psychological resilience—commands a massive premium, because it sells decision quality, not just information.

K-12 Classroom Teachers (Impact: Low / Net Positive) So far, the evidence points more toward task transformation than mass displacement. AI is increasingly being used for planning, assessment, feedback, and administrative work, while schools are also developing expectations around responsible and critical AI use. The result is not a new guarantee of job security, but a shift in the teacher’s role toward judgment, classroom management, student motivation, and helping learners use AI appropriately.

High-End, Bespoke 1-on-1 Tutoring (Impact: Minimal) This is the realm of scarcity premiums. The higher the stakes, the more customized the service, the less AI matters.

The rule governing this map is elegantly simple: The severity of the AI impact is inversely proportional to the distance from a standardized answer, and directly proportional to the distance from a human relationship. The more a job can be described in a standard operating procedure, the faster AI will hollow it out. The more a job relies on trust, nuance, and real-time emotional calibration, the more AI looks like a gift. To quote a veteran academic director interviewed for this piece: "Schools used to pay a flat rate for every part of the teaching process. Now, the market is forcing teachers to prove exactly which part of them is actually expensive."

VI. The Next 24 Months: Where the Pressure Will Land

If we project this three-tier model forward by two years, three specific pressure points emerge, along with one dangerous trap.

Signal 1: The Degree Reset In April 2026, Khan Academy, TED, and ETS announced the Khan TED Institute, with corporate partners including Google, Microsoft, Accenture, Bain, McKinsey, and Replit. The program is still under development and is pursuing accreditation, with the organizations designing an undergraduate degree intended to cost under $10,000 in total and targeting a 2027 launch. If that model reaches accreditation and proves viable, it could put new pressure on the pricing of standardized online degree programs. The more exposed segment would likely be programs whose value proposition depends heavily on scalable content delivery rather than on scarce forms of prestige, network access, or intensive human mentorship.

Signal 2: The Hollowing of Entry-Level Language Instruction Once AI voice synthesis and conversational latency cross the threshold of "good enough," the introductory market for high-resource languages (English, French, Spanish) will continue to contract. Duolingo’s aggressive strategy is essentially a land grab to ensure AI becomes the entry point for language learning before anyone else can claim it. This logic will inevitably bleed into every institution and freelance teacher whose primary income relies on beginner-level instruction. Intermediate and advanced language teaching will be insulated for longer; at that level, students are paying for cultural nuance, idiomatic expression, and human connection, not just vocabulary drilling.

The Trap: The "AI Content Reviewer" There is a specific type of transitional role that deserves caution: the "AI Output Reviewer." These are roles primarily focused on proofreading, tweaking, and verifying AI-generated lesson plans or educational content. They may be useful in the short term, but their long-term defensibility is uncertain because model quality and automated evaluation are likely to improve. Building an entire career around correcting routine AI errors therefore carries a substantial risk of role compression.

The Green Zones Conversely, some areas appear more resistant to near-term automation. Early childhood education involves physical care and emotional regulation; special education often requires highly individualized support; and high-stakes advising can depend heavily on trust, context, and accountability. These are not technology-proof occupations, but they contain more forms of value that are difficult to reduce to standardized outputs. AI fluency may also become increasingly useful for educators who can combine automation with visible improvements in personalization and service quality.

Predicting the future doesn't require mysticism. A more useful question is which parts of a job consist primarily of standardized output, and which parts depend on judgment, context, accountability, or relationships.

VII. The Operator’s Playbook: How to Survive the Repricing

For the individuals still navigating this industry, the time for abstract philosophy has passed. It is time for a tactical playbook. The core mandate is singular: You must reposition yourself from an "Answer Provider" to an "AI Workflow Conductor and Learning Experience Designer."

Here are five immediate actions to take, and three traps to avoid.

1. Audit Your Human-Machine Division of Labor Take a piece of paper and draw two columns. In the left column, write down everything you currently hand off to AI: generating quiz questions, grading rubrics, drafting diagnostic reports, creating slide decks. In the right column, write down what you must do yourself: the one-on-one check-ins with a struggling student, reading the emotional temperature of a room, the phone calls with anxious parents. You must audit this list every quarter. Why? Because the left column will naturally expand as AI capabilities improve. The true utility of this list is not just efficiency; it is a strategic compass. You must aggressively automate the left column so you can reinvest every freed-up hour into deepening the right column.

2. Make Your AI Workflow Visible (and Charge for It) Do not use AI just to save time behind the scenes; use it to create deliverables that justify a premium price. Use AI to generate five different difficulty tiers and three learning-style variations for a single math concept, and use your human expertise to curate the perfect mix for your student. Use AI to run an error-pattern analysis on a student's last three tests, identifying the psychological root of their mistakes. Deliver the pre-class diagnostic report directly to the parents, branded with your name and your pedagogical philosophy. When using AI becomes the industry baseline, "knowing how to use AI" is no longer a differentiator. "Using AI to engineer a level of personalization that a human couldn't do alone" is.

3. Productize "AI Literacy" Your students are already using AI. Ignoring that reality leaves an important part of their learning environment unmanaged. Teach them explicitly how to prompt, how to cross-verify AI-generated claims, and how to recognize confident errors. As schools and education systems develop expectations around responsible AI use, this can become an increasingly valuable part of an educator’s role. AI can help teach calculus; the educator’s contribution is to help students understand when to use AI, when to question it, and how to use it without outsourcing their own judgment.

4. Move Downstream if You Are in Content Creation If your primary job is generating raw educational content, you should reconsider where you create value. Do not compete with AI purely on production volume. The marginal cost of generating another worksheet or draft explanation can be extremely low, which makes routine production increasingly difficult to differentiate. Move downstream. Become the final arbiter of quality. Become the product designer who knows which content actually works for a specific demographic of learners. As AI lowers the barrier to production, taste, judgment, context, and accountability become more important sources of differentiation.

5. Avoid the Three Fatal Traps First, do not pretend AI doesn't exist. Parents and students are using it. If you don't define AI's role in your classroom, the internet will define it for you. Second, do not over-index on platform employers. The contract workers at Chegg received zero compensation and zero warning when their income went to zero. Build your own client roster and own your reputation. Third, do not anchor your career to "AI Reviewer" roles. (See Section VI).

The Ultimate Formula Your personal margin of safety in the education industry can be reduced to a simple equation: (Professional Judgment × AI Fluency) ÷ Standardizability

You cannot control the denominator—that is dictated by the nature of your job. Therefore, you must pour all your energy into the numerator. Cultivate a deeper mastery of your subject, refine your instructional design instincts, and master the AI tools that allow you to scale your empathy. A plausible high-value profile for the coming decade is a combination of deep subject-matter expertise, strong AI fluency, and the ability to build trust and connection with learners.

VIII. Epilogue

Let us return to the Ph.D. student in Ann Arbor. In the real world, there is no cinematic third-act redemption. He most likely moved on to a different gig, or focused entirely on his dissertation. The disappearance of his side income didn't even leave behind a memorable date. There was no final paycheck, no exit interview, no farewell email.

But that quiet, unceremonious exit is exactly what the education industry needs to remember. Some forms of educational work have not been replaced by a better way of teaching; they have simply become easier to reproduce through low-cost digital and AI systems. In those cases, demand for the original service can contract sharply rather than migrate neatly to a new human role.

At the same time, the work that remains may be valued differently. Judgment, trust, accountability, and the ability to keep another human being engaged in the hard work of learning are difficult to reduce to a standardized output. As AI makes answers cheaper, these forms of human contribution may become more economically important.

For those still standing in the industry, this is both a profound threat and a rare, transparent window for career reinvention. But windows like this do not stay open forever. The graduate student staring at his Chegg dashboard two years ago, watching the queue slowly empty, had no idea he was looking at the most honest signal the market had ever sent. The answer economy is dead. The relationship economy is just opening for business.

References

Chegg, Inc., 2026 2025 Annual Report https://www.sec.gov/Archives/edgar/data/1364954/000136495426000042/a2025annualreport.pdf
Oreopoulos, Philip and Nina Low, 2026 One Click Away: AI Tutoring with Khanmigo in a Two-Year School Experiment https://www.nber.org/papers/w35620
College Board, 2025 New Research: Majority of High School Students Use Generative AI for Schoolwork https://newsroom.collegeboard.org/new-research-majority-high-school-students-use-generative-ai-schoolwork