Is AI Already Breaking the Economics of Instructive Teaching?

From coaching economics to the changing economics of white-collar expertise

1. The Question Is Bigger Than Teaching

For decades, education has operated on a simple economic constraint: instruction requires human time. A teacher can teach one class, tutor one student, review a limited number of assignments, and answer a limited number of questions. Even the best teacher eventually reaches the same physical constraint as everyone else: there are only so many hours in a day.

AI changes the economics of that constraint. An AI system can explain the same concept repeatedly, adapt explanations, generate examples, answer questions, evaluate responses and provide feedback without consuming another human hour for every additional student.

That raises a question that goes beyond education: what happens to the price of expertise when the amount of human time required to deliver it falls dramatically?

Teaching may simply be one of the first places where the question becomes visible.

2. When Instruction Stops Being Scarce

The traditional coaching business is built around scarcity. A good teacher has limited hours, a good classroom has limited seats, and a premium institution therefore sells access to scarce instructional capacity.

Online education has already weakened this model. A recorded lecture can reach thousands of students. Digital study material can be reproduced with almost no additional distribution cost. Large online platforms can spread the cost of producing instruction across enormous numbers of learners.

AI introduces another step. It does not merely distribute a teacher’s material more cheaply; it can potentially perform part of the instructional process itself.

The economic progression is therefore significant. Education moves from instruction tied to a teacher’s physical time, to digitally distributed instruction, and potentially toward instruction that can be generated and delivered by computational systems at very large scale.

The crucial question is no longer simply how many students one teacher can teach. It is how many students one instructional system can serve.

3. The 918× Question

The StudentBench study discussed in the previous article provides a useful trigger for this economic question.

In its reported GRE tutoring experiment, the researchers compared AI tutoring with expert human tutoring. The reported learning gains were statistically comparable in the experiment, while the calculated cost per percentage point of improvement was approximately $0.0052 for the AI tutor versus $4.81 for the human tutor. That produced a reported cost difference of roughly 918×.

This does not mean that AI tutoring is universally 918 times better than human tutoring. The study was specific to its setting, intervention and measurement, and it does not establish that teachers can universally be replaced.

But the economic question remains powerful. What happens to a market when a sufficiently comparable unit of instructional output can potentially be produced at radically different costs?

That question is more important than the particular 918× number.

4. From AI Capability to Market Pressure

Capability by itself does not necessarily destroy an economic model. Cost changes can.

Suppose two systems can deliver sufficiently similar outcomes for a particular task, but one requires a large amount of human labour while the other requires relatively little. The market eventually begins asking a simple question: why should the old price continue to exist?

The cheaper system does not have to be perfect. It only needs to be good enough for a sufficiently large portion of the market.

This is one of the characteristic ways technological disruption begins. The new technology does not initially have to eliminate the old system. It can simply make parts of the old cost structure increasingly difficult to justify.

The evidence is moving beyond the 2025 discussion. In May 2026, Pearson reported new data from higher-education students using AI-powered adaptive practice. Students using the adaptive system were 90% more likely to reach initial mastery than students using traditional static practice, without spending more study time. Pearson’s researchers describe this as evidence that AI can improve learning outcomes when it is designed around the learning process rather than simply generating answers.

This matters for the economic argument because it provides a more recent example of AI moving from assistance toward measurable instructional performance. The question is no longer only whether AI can make a teacher’s preparation faster. Increasingly, the question is whether AI can perform parts of the learning interaction itself—and do so at a scale and cost structure that human instruction cannot easily match.

The distinction is important. Pearson’s 2025 education report emphasized pedagogy, responsible use and AI working alongside educators. Its newer 2026 research shows that the capability is also being measured in actual student learning. The two ideas are compatible: AI can augment teachers while simultaneously changing the economics of the instructional tasks it performs.

5. What Is the Student Actually Paying For?

A coaching fee is not really payment for “teaching” alone. It is payment for a bundle of services that may include explanation, examples, doubt solving, testing, feedback, diagnosis of mistakes, study planning, motivation, discipline, peer environment, accountability, reputation and parental confidence.

AI can potentially attack several components of this bundle. Explanation, practice, testing, feedback and repetitive doubt solving are particularly amenable to computational systems.

Other components are more difficult to commoditize. Motivation, personal accountability, mentoring, social environment and human trust may continue to have substantial value.

This suggests a useful conceptual model. A large portion of routine instructional activity could increasingly become AI-deliverable, while a smaller but economically important layer remains strongly human.

The 80:20 split is an illustrative model, not an empirical measurement. Its purpose is to decompose the coaching product and ask which parts are genuinely scarce when AI becomes capable of performing routine instruction.

6. The Teacher’s Hidden Product

The most interesting thing a great teacher possesses is not simply information. It is methodology.

How does the teacher recognize that a student has misunderstood something? Which example does the teacher choose? In what sequence should the concepts be introduced? What question should come next? How does the teacher distinguish a careless mistake from a conceptual misunderstanding?

And perhaps most importantly, how does the teacher decide what the student needs next?

This is the teacher’s hidden product.

That connects directly to the Soft Brain idea developed in the first article. The valuable asset is not merely a collection of facts. It is the expert’s method of thinking: how the expert selects examples, diagnoses mistakes, sequences concepts, asks questions and decides the next step.

Once that methodology can be encoded into an AI system, the expert’s knowledge is no longer constrained entirely by the expert’s personal calendar.

The expert does not disappear from the system. The expert becomes the source and curator of the methodology.

7. From Selling Hours to Scalable Expertise

The traditional expert business model has an unavoidable bottleneck. The expert sells limited hours, and therefore the economic output of the expertise remains tied to the expert’s available time.

AI potentially changes that relationship.

A teacher who personally teaches 500 students has one economic limit. A teacher whose methodology becomes part of an AI tutoring system could potentially influence a vastly larger number of students without personally repeating every explanation.

This is the transition from selling hours to building scalable expertise.

The scarce asset may therefore shift. Instead of the expert’s time being the primary constraint, the quality of the expert’s cognitive system could become more important.

That was the conceptual proposition of the first article. The second article then asked whether AI can actually reproduce meaningful portions of expert performance. This article asks what happens when that capability enters a market.

8. Why Instructive Teaching Is Especially Exposed

Teaching contains many activities that are structurally suitable for AI. A system can explain a concept, demonstrate a method, ask a question, evaluate an answer, identify an error, generate another example and repeat the process.

The important characteristic is that much of this activity can be represented as information processing.

A student asks a question and the system responds. The student attempts a problem and the system evaluates it. The student makes a mistake and the system attempts to identify the underlying misconception. The system then generates another explanation or another problem.

The cycle can continue for as long as the student needs it.

For a human tutor, every additional cycle consumes another piece of scarce human time. For an AI system, the marginal cost of another interaction can be dramatically lower.

That is the economic vulnerability.

9. But Teaching Is More Than Instruction

This is also why the simple statement “AI will replace teachers” is too crude.

A teacher may be an instructor, but may also be a mentor, judge, motivator, role model and source of accountability. A student preparing for a difficult examination does not necessarily need only an answer.

Sometimes the student needs someone to recognize a recurring behavioural pattern and intervene. A teacher may notice that a student understands the mathematics but is avoiding difficult problems, or that the student is repeatedly making the same conceptual mistake.

Those are not merely information-transfer problems. They involve judgment, motivation, relationships and responsibility.

The economic question therefore becomes more interesting than “human teacher or AI?” It becomes a question of which parts of teaching should be performed by AI and which parts should remain human.

10. The New Division of Labour

A plausible division of labour could emerge in which AI increasingly handles explanation, practice, testing, feedback and repetition, while humans concentrate more heavily on direction, judgment, motivation, accountability and mentoring.

This would not necessarily mean that the teacher disappears. It would mean that the composition of the teacher’s job changes.

The worker remains, but the pricing model changes.

A teacher may therefore spend less time delivering routine explanations and more time deciding what the student should learn, interpreting unusual difficulties, motivating the student and taking responsibility for the overall learning process.

The value of the human contribution could move upward in the instructional hierarchy.

11. What Is a Coaching Company Actually Selling?

This is where the business question becomes uncomfortable.

If AI can provide increasingly capable explanations, unlimited practice, instant feedback and individualized interaction at very low marginal cost, what exactly is a coaching company charging for?

It may still be selling brand, selection, testing, physical environment, peer competition, discipline, mentorship, reputation and results. Those elements do not automatically disappear because an AI tutor becomes capable.

But routine instruction may become a weaker source of scarcity.

That could fundamentally change the economics of the coaching company. The business may gradually move from selling access to teachers toward selling an integrated learning system in which AI performs a large instructional layer and humans provide the elements that remain difficult to automate.

12. When Better Instruction Becomes Cheaper

There is an unusual possibility here. Technology can simultaneously make instruction better and cheaper.

We normally associate higher quality with higher price because better human expertise is scarce. A highly regarded teacher has limited time, and scarcity allows that expertise to command a premium.

AI can break part of that relationship.

As the underlying system becomes more capable, the cost of providing another unit of instruction can simultaneously decline.

This creates an unusual economic combination: capability can rise while cost falls.

That is one of the reasons AI is potentially different from simply hiring more teachers. Hiring more teachers increases capacity, but it does not fundamentally change the relationship between instructional output and human labour.

AI potentially changes that production function.

13. The Price of Expertise Can Fall Even While Capability Rises

This may be the most important economic principle in the entire argument.

People often assume that if expertise becomes more valuable, experts must become more expensive. That is generally true when expertise remains scarce.

But AI changes the scarcity equation.

Suppose an expert can produce a certain amount of valuable cognitive work in ten hours. If an AI system can reproduce much of the expert’s methodology and perform millions of similar interactions, the capability associated with that expertise may increase even as its scarcity decreases.

That creates a paradox.

The price of expertise can fall even while the capability of expertise rises.

The distinction is between the quality of intelligence and the scarcity of access to that intelligence. AI may improve the first while reducing the second.

This distinction becomes critical when we move beyond education.

14. The Education Price-Compression Cycle

The mechanism can be understood as a cycle. Increasing AI capability lowers the cost of some instructional activities. Lower costs change student expectations. Those expectations create price pressure on traditional providers. Providers respond with new business models, which change the requirements for human labour and encourage further adoption of AI.

The cycle can therefore become self-reinforcing.

There is also an important reason to look at the coaching industry before claiming that AI alone is responsible for the current price pressure.

Vijay Jha, writing from his own experience in the Indian coaching industry, recently described branded JEE/NEET programmes as having cost roughly ₹90,000–₹1.4 lakh before COVID, while observing programmes around ₹40,000 in the current market.

His observation should be treated as first-person industry experience, rather than as a nationwide statistical measurement. Indian coaching-faculty salaries and compensation are also not reported through a comprehensive public dataset comparable to some Western labour markets.

Nevertheless, the observation illustrates an important phenomenon: substantial price compression was already occurring before AI had fully entered the economics of instruction.

Online platforms made instruction cheaper to distribute.

AI potentially makes portions of instruction cheaper to produce.

Those are two different stages of the same economic transition.

15. From Human Time to Computational Capacity

And now the argument becomes much bigger than education.

The same economic mechanism is visible in other forms of white-collar work. Software engineering, coding, research, analysis, writing, design, customer support, marketing and several other cognitive occupations contain tasks that can increasingly be performed or accelerated by AI systems.

Consider software engineering. Traditionally, a programmer converts knowledge and time into software. AI increasingly allows the programmer to combine knowledge with computational assistance to produce more output in the same period.

The important question is not whether every software engineer disappears.

It is: how much human time is required to produce one unit of economically valuable cognition?

That may become one of the most important economic variables of the AI era.

16. The White-Collar Echo

This is why the anxiety visible across professional networks is broader than the coaching industry.

LinkedIn, X and Reddit contain large numbers of discussions about layoffs, hiring difficulties, changing software-engineering roles, AI-assisted coding and the changing value of previously scarce skills.

These platforms should not be treated as statistical measurements of the entire labour market. Their populations are highly selected, and highly visible negative experiences can distort perception.

But they are useful as social sensors.

They allow us to observe something that formal labour statistics may take considerably longer to capture: how workers themselves are experiencing a rapidly changing definition of valuable work.

The question appearing across these discussions is remarkably similar: if AI can perform part of what I was paid to do, what exactly remains scarce about my labour?

That question appears in software engineering. It appears in writing, design, research and analysis. It appears in teaching.

The occupations are different, but the underlying economic problem is similar.

17. What Happens to the Human Expertise Premium?

If routine cognitive work becomes increasingly abundant, scarcity does not necessarily disappear. It can move to another layer.

Human value may increasingly attach to judgment, trust, responsibility, context, reputation, relationships, direction and the ability to decide what should be done rather than merely how it should be done.

This could produce a strange reversal. The most valuable human contribution may increasingly occur above the layer where AI becomes extremely capable.

A human may decide what problem is worth solving, while AI explores thousands of ways to solve it. The human may decide which outcome is acceptable, while AI generates possible solutions. The human may accept responsibility for the decision, while AI provides the computational capacity.

The new scarcity may therefore be direction rather than knowledge.

18. From the Economics of Teaching to the Economics of Intelligence

Now the three articles connect.

The first article asked whether human expertise could become scalable. Its conceptual answer was that human methodology could potentially become part of a scalable cognitive system.

The second article asked whether AI can reproduce meaningful portions of expert performance. The StudentBench evidence provided one concrete example of that possibility in tutoring.

The third article asks what happens when scalable, machine-mediated expertise enters an economy built around selling human cognitive labour by the hour.

That is why teaching is only the beginning.

The same transition can potentially move through software engineering, design, analysis, research, consulting and many other forms of knowledge work. The common denominator is not the profession. It is human cognitive time.

For centuries, economically valuable intelligence was tightly coupled to the person who possessed it. If you wanted the expert’s knowledge, you generally had to obtain some portion of the expert’s time.

AI is beginning to loosen that coupling.

Once intelligence becomes partially separable from the human hour, the economics of white-collar work can change.

19. What Happens When Instruction Is Almost Free?

Imagine a student who can ask an AI tutor unlimited questions. The tutor never gets tired, can explain the same concept in many different ways, remembers the student’s mistakes, generates another problem immediately and adjusts the difficulty according to the student’s performance.

If the cost of another interaction becomes extremely small, the economics of instruction changes fundamentally.

The question is no longer simply whether AI can replace the teacher.

The more interesting question is: if instruction becomes almost free, what remains uniquely valuable about a human teacher?

Perhaps it will be motivation. Perhaps judgment, trust or accountability. Perhaps the ability to understand the human being behind the question. Perhaps teachers will become more valuable precisely because routine instruction becomes abundant.

We do not yet know exactly where that boundary will settle.

But the economic direction is becoming clearer.

The transformation may not be about replacing human expertise with artificial expertise. It may be about separating intelligence from the scarcity of human time.

And once that happens, the question is no longer confined to teaching.

It becomes a question about the economics of intelligence itself.

Read the Series