Why the Most Counterintuitive Economic Idea of the 19th Century May Define the 21st
Introduction
Businesses don’t win by getting smaller. They win by investing, innovating, and growing. The organizations that rise to the top are those that see technology not as a threat to their headcount, but as an engine that amplifies what their people can do.
Yet every time a powerful new technology arrives — the steam engine, the spreadsheet, the internet, and now artificial intelligence — the same anxious question echoes across boardrooms and newsrooms: Will this take our jobs?
This article argues that the most revealing answer to that question doesn’t come from Silicon Valley. It comes from a dusty 19th-century economics text about coal.
Part I: The Prophecy and the Reality Check
The Godfather Speaks
In 2016, Geoffrey Hinton — the British-Canadian computer scientist so foundational to modern AI that he is widely called the “Godfather of AI” — made a dramatic prediction. “People should stop training radiologists now,” Hinton declared. “It’s just completely obvious that in five years, deep learning is going to be better than radiologists.”
This statement landed with the weight of authority. Hinton’s work on deep neural networks genuinely did lay the intellectual foundation for much of contemporary AI. If anyone knew where the technology was heading, it was him.
The prediction, however, has not aged well.
Ten years on, not only have radiologists not gone extinct — we are actually training more of them than when Hinton made that statement. AI is performing extraordinary feats in medical imaging: detecting tumors earlier, flagging anomalies more consistently, reducing diagnostic error in carefully studied scenarios. And yet the profession has grown. The radiologist’s role has evolved — integrating AI tools, focusing on complex cases and patient communication, supervising algorithmic outputs — but it has not vanished.
Why? The answer is not that AI failed. The answer is something much more interesting, and much older than any silicon chip.
Part II: A Victorian Economist and the Counterintuitive Truth
William Stanley Jevons and the Coal Question
In 1865, a young English economist named William Stanley Jevons published a book called The Coal Question. England was the coal-burning capital of the world, and the nation’s leaders were worried: with James Watt’s newer, more fuel-efficient steam engines spreading across industry, would Britain burn through its coal reserves faster — or slower?
The intuitive answer seemed obvious: more efficient engines burn less coal per unit of work, therefore total coal consumption will fall.
Jevons argued the opposite. He spent an entire chapter making the case that greater efficiency in coal-powered technology would — paradoxically — lead to more consumption of coal, not less.
His logic was precise. More efficient steam engines lowered the effective cost of energy. Lower cost made coal power economically viable for entirely new industries and applications that had previously been impractical. More industries adopted coal. Total demand exploded. Britain didn’t burn less coal because of Watt’s engine — it burned dramatically more.
“It is wholly a confusion of ideas,” Jevons wrote, “to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth.”
This insight became known as Jevons Paradox: when a technology makes a resource more efficient to use, total consumption of that resource often increases rather than decreases, because lower costs unlock new demand and entirely new use cases.
Part III: The Paradox Has a Perfect Track Record
The ATM That Didn’t Kill Bank Tellers
In 1967, the world’s first ATM was installed in London. The machines could dispense cash faster, more cheaply, and more accurately than any human teller. Predictions of mass layoffs in the banking sector spread widely.
What actually happened? ATMs reduced the cost of running a branch. Lower branch costs meant banks could profitably open more branches — urban bank locations increased by roughly 43% over the following two decades. More branches meant more tellers. Teller employment actually rose. The job didn’t disappear; it evolved from mechanical cash-handling toward relationship banking and complex financial advisory work that machines couldn’t do.
The Spreadsheet That Made Accountants Multiply
When VisiCalc — the first consumer spreadsheet program — launched in 1979, it could perform in seconds the arithmetic calculations that had occupied a skilled accountant for an entire day. Around 400,000 bookkeeping positions eventually disappeared as a result.
But something else happened simultaneously: 600,000 new accountant positions were created. The U.S. Bureau of Labor Statistics counted roughly 339,000 accountants in 1980. By 2022, that figure had grown to approximately 1.4 million. Spreadsheets didn’t reduce demand for financial expertise — they made financial analysis so cheap that every business wanted vastly more of it. Accountants who once labored over arithmetic were freed to do higher-level financial modeling, strategic analysis, and forecasting. Entirely new categories of work emerged.
The Smartphone That Invented Professions
When the smartphone arrived, not a single serious forecaster predicted the app economy. No one foresaw that a device for making calls would conjure professions that had no name and no precedent: UX designers, mobile platform engineers, social media managers, gig economy coordinators, app store optimization consultants, influencer marketing strategists. The phone didn’t shrink the labor market. It invented new continents within it.
The Pattern Is Clear
Each time, the same story: a tectonic innovation makes something radically cheaper, triggers panic about displacement, and then — almost embarrassingly — ends up raising productivity, expanding demand, and lifting far more boats than it sinks. Electrification turned factories into engines of mass employment. The PC and spreadsheets “eliminated” accountants, only to create entire finance and software industries. The internet hollowed out some media jobs while spawning e-commerce, digital marketing, cloud computing, and the modern startup ecosystem.
As Microsoft CEO Satya Nadella wrote in early 2025, when Chinese AI company DeepSeek built a cutting-edge model at a fraction of the expected cost: “Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket.”
Part IV: AI and Jevons — The Connection
The New Coal Is Intelligence
AI is not coal, of course. But structurally, what AI does to cognitive labor is exactly what Watt’s steam engine did to physical energy: it makes it radically cheaper and more scalable.
Before AI, complex analysis, personalized content generation, nuanced customer service, legal research, and medical diagnosis required expensive, highly trained human specialists. AI dramatically reduces the cost of performing these cognitive tasks. And — precisely as Jevons would predict — that cost reduction does not simply reduce the number of people employed to do these things. It makes cognitive work economically viable at scales and in applications that were previously impractical.
The binding constraint on human knowledge work has shifted. It is no longer the execution of a task that is expensive — AI absorbs execution at scale. The constraint is now judgment, specification, oversight, accountability, and the ability to own outcomes. That is where humans become more essential, not less.
What Jevons Predicts for AI Employment
Applied to AI, Jevons Paradox generates a specific prediction: AI reduces the labor required per task, but dramatically expands what is economically feasible to attempt. This increases the total amount of work humans are employed to support.
This is already visible in the data. AI and machine learning specialist roles grew 143% year-over-year in demand in 2025. Not because companies are adding overhead, but because lower costs of execution are unlocking entire categories of work that weren’t viable before.
The World Economic Forum’s Future of Jobs Report 2025 — drawing on surveys of over 1,000 leading global employers representing more than 14 million workers across 55 economies — projects that shifting technology trends will generate 170 million new jobs globally by 2030, while displacing 92 million others. The net result: a gain of 78 million positions. That is a 7% net increase in total formal employment, driven substantially by AI.
Part V: The New Map of Work
Jobs That Will Disappear
Jevons Paradox does not mean no jobs disappear. It means the aggregate outcome is expansionary, not contractionary. Routine, low-discretion roles are genuinely at risk: data entry clerks, basic bookkeepers, assembly-line quality inspectors, formulaic content producers. These are the “elevator operator” situations — where the demand for a specific task is fixed, and AI fully substitutes rather than complements human effort.
The transition pain is real. Job losses in these categories tend to be sharp and localized, while Jevons-style job creation unfolds more slowly and often in different geographies and occupational categories. That temporal and geographic gap is where real human hardship lives, and it demands serious policy responses around reskilling, income support, and regional economic development.
Jobs That Will Grow — and Why
The new map of work is defined by what AI cannot absorb. Genpact CEO BK Kalra has put it this way: frontier AI can handle roughly 80% of a business process effectively. The remaining 20% — edge cases, regulatory nuance, decisions that don’t follow patterns, situations requiring genuine accountability — is where actual business value concentrates.
That 20% is human territory.
1. Entirely New Job Categories
Just as smartphones created professions that didn’t exist before 2007, AI is creating entirely new roles: AI product managers, AI safety engineers, prompt engineers, model auditors, AI ethics officers, alignment researchers, and synthetic data specialists. These jobs are artifacts of the technology itself — they exist because the AI requires human stewardship that no prior technology demanded in the same form.
2. Expansion of Long-Tail Services
Perhaps the most socially significant consequence of Jevons dynamics in AI is the expansion of services that were previously available only to the wealthy. Custom tutoring was once accessible to families who could afford private tutors at $80–$150 per hour. AI brings personalized, adaptive learning within reach of anyone. Niche legal analysis — say, a detailed review of your freelance contract’s intellectual property clauses — required expensive specialized attorneys. AI makes that analysis affordable for ordinary workers. Personalized healthcare support, detailed financial planning, bespoke mental health coaching: services long rationed by price may become widely accessible. This democratization of expertise creates its own expansion of demand.
3. High-Context, High-Accountability Roles
The roles that grow most robustly are those requiring what might be called contextual judgment: the ability to read ambiguous situations, apply institutional knowledge, navigate regulatory nuance, and own the consequences of decisions. These include:
- Problem framers and goal setters — people who can define what the AI should be doing, which is harder than it sounds
- AI supervisors and evaluators — humans who verify that AI outputs are accurate, fair, and appropriate in context
- Governance and compliance officers — as AI-generated content and decisions proliferate, the infrastructure of oversight, auditability, and regulatory compliance must scale with it
- Trust-critical customer-facing roles — there is a category of human interaction where the customer, patient, or citizen needs to feel genuinely heard by a human being. Healthcare, legal counsel, mental health, financial planning during a life crisis: these interactions have emotional stakes that pure automation cannot serve
- Cross-functional coordinators — as organizations become more complex and AI-augmented, the ability to bridge between technical systems, business strategy, regulatory requirements, and human teams becomes more valuable
4. Integration, Oversight, and Quality Assurance
Deploying AI at scale is not a “set it and forget it” exercise. It requires ongoing human investment in integration, monitoring, quality control, and trust-building. BCG data shows that 74% of enterprises struggle to scale AI value — and that is primarily a people and process problem, not a technology problem. Leaders allocate 70% of their AI-related effort to people and process challenges. This is a large and growing category of human work.
Part VI: The Organizational Divide
The Cutting Mistake
There is a predictable failure mode for organizations navigating AI: treating it purely as a cost-cutting instrument. Reduce headcount, shrink the workforce, trim the payroll. The logic is seductive in a spreadsheet. It is catastrophic in practice.
The organizations that win with AI are not the ones that use it to get smaller. They are the ones that use it to get smarter and larger. They leverage AI-driven efficiency gains to enter new markets, serve previously underserved customers, move faster on product development, and raise the ceiling of what is achievable with their existing teams.
Smart companies understand Jevons Paradox. They recognize that AI will require more employees, not fewer — but employees operating at a higher level of skill, judgment, and creativity.
The not-so-smart organizations will be so focused on reducing the number of cars in the employee parking lot that they will miss the train already pulling away from the station. They will cut their way to mediocrity while their competitors innovate their way to dominance.
The Augmentation Imperative
The winning framing is augmented intelligence, not artificial intelligence. AI as a force multiplier for human capability, not a replacement for it. A radiologist augmented by AI can review more cases, catch more edge cases, and spend more clinical time on complex diagnosis and patient communication. A financial analyst augmented by AI can model more scenarios, identify more risks, and provide more nuanced strategic counsel. A teacher augmented by AI can personalize instruction for every student simultaneously while focusing their own attention on motivation, mentorship, and social-emotional learning.
In every case, the human doesn’t disappear. The human becomes more valuable.
Part VII: The Skills for the AI Era
What Smart Employees Will Need
The workforce of the AI era is not primarily a technical workforce. The WEF’s Future of Jobs Report 2025 identifies the fastest-growing skills by 2030 as a combination of technological literacy and deeply human capabilities: creative thinking, resilience, flexibility, agility, curiosity, and lifelong learning. Eight of the top ten most in-demand skills identified in major workforce research are classified as durable human skills — not readily automatable.
1. Adaptability and Flexibility
The AI landscape changes at a pace that makes any fixed skillset precarious. The professionals who thrive will be those who can update their mental models rapidly, embrace new tools without anxiety, and adjust their approaches as the technology evolves. Adaptability is not a soft skill in the AI era — it is the core competency.
2. Lifelong Learning
The half-life of specific technical knowledge is shortening dramatically. Staying current is not a one-time effort but a continuous practice. Workers who treat learning as an ongoing habit — reading, experimenting, taking courses, following new developments — will have a compounding advantage over those who view education as something that ended with their degree.
3. Critical Thinking
AI is powerful but not infallible. Models hallucinate facts with confident fluency. They optimize for specified objectives in ways that may diverge from what was actually intended. They reflect biases present in their training data. Human critical thinking — the ability to question AI outputs, verify claims, identify gaps in reasoning, and evaluate whether the AI is doing what you actually want — is not a nice-to-have. It is the essential check on a powerful but imperfect system. The human in the loop must retain genuine decision-making authority, not simply rubber-stamp what the model produces.
4. Creativity and Problem Framing
One of AI’s most significant gifts to human workers is time. By automating the routine, the repetitive, and the mundane, AI frees human attention for higher-order thinking. People who can use that reclaimed time well — who can think creatively, identify unconventional solutions, frame problems in new ways, and imagine applications that don’t yet exist — will find themselves more valuable, not less.
5. Emotional and Social Intelligence
Trust is not automatable. Empathy is not a model output. The ability to read a room, navigate conflict, motivate a team, build relationships across cultural differences, and provide genuine human presence in emotionally charged moments remains stubbornly resistant to technological substitution. These skills become more differentiating as AI handles more cognitive execution.
6. AI Fluency
Workers who understand how to work with AI effectively — who know how to prompt well, evaluate outputs critically, integrate AI tools into workflows, and communicate clearly about AI capabilities and limitations — earn a measurable premium. PwC’s 2025 AI Jobs Barometer found that workers with demonstrable AI skills earn on average 25% morethan peers without them. AI fluency is rapidly becoming as fundamental as digital literacy was in the early 2000s.
Part VIII: The Numbers Behind the Narrative
The empirical case for Jevons dynamics in AI is accumulating:
- The WEF Future of Jobs Report 2025 projects 170 million new roles created by 2030 against 92 million displaced — a net gain of 78 million jobs
- AI/ML engineer roles grew 143% year-over-year in 2025, driven not by overhead expansion but by new use cases becoming economically viable
- Hybrid roles — positions combining domain expertise with AI competency — are projected to grow from under 200,000 in 2010 to 2.91 million by 2035
- 85% of employers plan to prioritize workforce upskilling; 77% aim to upskill staff specifically for AI-adjacent work
- 41% of employers plan to reduce headcount through AI automation, but 47% plan to transition affected workers into other internal roles — suggesting even organizations pursuing efficiency gains are investing in workforce continuity
- Industries with higher AI adoption show 4x higher productivity growth than less AI-intensive sectors
- Goldman Sachs research notes that 60% of U.S. workers today are employed in occupations that did not exist in 1940 — an extraordinary testament to technology’s job-creation capacity over time
Part IX: The Caveats — What Jevons Does Not Guarantee
Intellectual honesty requires acknowledging where Jevons Paradox has limits.
The elevator operator problem is real. When elevators became automated, elevator operator jobs disappeared entirely. Demand for elevator rides was fixed — you can’t have more rides just because operators are more efficient. When demand for a task is capped, efficiency simply means fewer workers. For roles that are fully substitutable by AI with no elastic demand upside, displacement is permanent.
Transition costs are severe and unevenly distributed. Even in optimistic aggregate scenarios, the path from here to an AI-abundant economy runs through a valley of disruption. Workers in displaced roles face real hardship: retraining time, geographic relocation, age discrimination, financial vulnerability during career transition. The people who bear these costs are often those with the fewest buffers. The fact that the aggregate economy expands does not automatically mean that the specific truck driver, data entry clerk, or paralegal displaced by AI gets a share of that expansion.
The speed of change matters. Historical technological transitions — the agricultural revolution, industrialization, electrification — unfolded over decades, giving labor markets time to absorb change. AI may compress this timeline in ways that outpace natural workforce adaptation. The risk is not permanent unemployment — it is a painful transition gap that requires active policy intervention to bridge.
Distribution is not guaranteed. Jevons Paradox predicts aggregate expansion of activity and employment. It says nothing about how those gains are distributed. If AI-era productivity accrues primarily to capital holders and a small technical elite, the aggregate employment number may look healthy while median worker welfare deteriorates.
Conclusion: The Train Is Already Leaving the Station
The core insight of this article can be stated simply:
Efficiency gains do not reduce demand for human work. They expand what work is worth doing.
This was true for coal in 1865. It was true for spreadsheets in 1979. It was true for the internet in the late 1990s. It appears to be true for AI now.
Smart organizations understand this. They are not asking “How do we replace our employees with AI?” They are asking “How do we use AI to make our employees capable of things they couldn’t do before? What new markets can we enter? What services can we offer that were previously cost-prohibitive? What competitive advantages can we build?”
Smart employees understand this too. They are not waiting for the disruption to arrive. They are building adaptability, developing AI fluency, deepening their critical thinking, and investing in the irreducibly human skills — creativity, judgment, empathy, accountability — that AI amplifies rather than replaces.
The organizations and individuals who grasp Jevons Paradox will ride the wave of AI-driven expansion. Those who don’t — who see only the threat, who focus only on cutting costs, who mistake the efficiency gain for the end of the story — will find themselves watching from the platform as a very fast train disappears over the horizon.
The future belongs to augmented intelligence. Not artificial intelligence. Not diminished human intelligence. Augmented.
The question is not whether AI will change the nature of work. It will, profoundly. The question is whether you will be the kind of organization, and the kind of professional, positioned to benefit from that change.
The answer to that question starts now.
Sources: World Economic Forum Future of Jobs Report 2025; PwC AI Jobs Barometer 2025; BCG AI Adoption Research; McKinsey State of AI 2025; Genpact/Newsweek; NPR Planet Money; Harvard Business Review; LinkedIn Workforce Report; U.S. Bureau of Labor Statistics.
How Smart Organizations Will Use AI: Jevons Paradox and the Future of the Workforce was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.