AI will displace tasks but also generate new domains of work as cheaper intelligence expands demand—a dynamic explained by Jevons’ Paradox. The article examines layoffs, role redesign, and emerging opportunities in AI governance, education, auditing, entrepreneurship, and human-centred services. It also warns that restricted access to advanced AI systems may widen knowledge gaps between developed and developing nations, making inclusive policy, infrastructure, and AI literacy urgent priorities.
AI Incursion
Artificial intelligence is often blamed for “invading” the workplace. The metaphor captures the speed and unease of the moment, but it can also mislead. AI does not enter an occupation like an army entering a territory, replacing every worker at once. It enters through specific tasks: drafting a contract, interpreting a scan, answering a customer query, debugging software, assessing credit or preparing a lesson. The occupation may survive even as its internal architecture is radically altered.
The all-important question, therefore, is not whether AI will destroy or create jobs in the abstract. Rather, it is who will lose which tasks, who will take on new responsibilities, how quickly institutions can adjust, and whether access to the most capable systems will be sufficiently widespread to prevent a new international hierarchy of knowledge.
Labour-saving Tech to Labour-reorganising Tech
AI can affect employment through three distinct mechanisms.
First, it can substitute for labour by performing tasks previously assigned to people. Routine documentation, basic coding, transcription, translation, scheduling, customer support and preliminary research are already susceptible to this form of automation.
Second, it can augment labour. A doctor may use AI to examine medical evidence without surrendering clinical judgement; a teacher may generate differentiated learning materials while retaining responsibility for pedagogy; an entrepreneur may conduct market analysis that was previously affordable only to large companies.
Third, AI can create entirely new tasks, products and markets. This “reinstatement effect” is often overlooked because jobs eliminated today are visible, whereas occupations that will emerge tomorrow do not yet have familiar names.
Likely areas of employment growth include AI auditing, model evaluation, algorithmic accountability, synthetic data management, cybersecurity, provenance verification, human–AI interface design, digital public infrastructure, and sector-specific AI integration. Healthcare will require professionals who can translate machine recommendations into clinically and ethically defensible decisions. Education will need AI curriculum designers, assessment-integrity specialists and mentors capable of rebuilding curiosity and independent thought in highly automated learning environments. Law, finance and public administration will require professionals who can determine when an algorithmic decision is valid, contestable or discriminatory.
Other opportunities will arise precisely because machines remain deficient in embodiment, accountability and human trust. Care work, counselling, negotiation, community leadership, complex sales, conflict resolution and skilled trades may become more valuable—not because they are untouched by AI, but because AI increases the relative scarcity of credible human presence.
Jevons’ paradox: Why Greater Efficiency Can Produce More Work
The most intriguing argument for employment expansion comes from Jevons’ paradox, propounded by the British economist William Stanley Jevons, who examined coal consumption in his 1865 book The Coal Question. Jevons observed that James Watt’s more efficient steam engine did not reduce Britain’s aggregate consumption of coal. By making steam power cheaper and commercially useful in more settings, efficiency expanded its applications so dramatically that total coal use increased.
Its original purpose was to challenge the assumption that technological efficiency would conserve a scarce resource. In modern economic language, the paradox occurs when an efficiency improvement lowers the effective price of a service, causing demand to rise by more than the initial saving. A rebound below 100 per cent merely reduces the expected saving; a rebound exceeding 100 per cent produces the full Jevons effect.
Applied to AI, the scarce “resource” is not necessarily labour itself but economically usable intelligence. If AI reduces the cost of producing analysis, software, design, diagnosis or personalised instruction, society may demand much more of these services.
When software becomes cheaper to produce, firms may not simply employ fewer programmers; thousands of smaller organisations may commission software they could never afford before. If AI lowers the cost of preliminary legal analysis, individuals and small enterprises may seek more legal assistance. If personalised tutoring becomes inexpensive, education may shift from occasional standardised instruction to continuous learning support. Lower costs could similarly expand preventive medicine, local-language publishing, scientific simulation, financial advice and services for people with disabilities.
Consequently, employment can rise in complementary activities: defining problems, acquiring clients, checking outputs, integrating systems, exercising judgement, and assuming responsibility. AI may perform more intellectual work while simultaneously expanding the market for human-guided intellectual services.
This is the hopeful interpretation of Jevons’ paradox: efficiency not only compresses existing production; it can release previously latent demand.
Why Jevons is not a Law of Employment
The paradox, however, should not be converted into technological optimism. Coal is consumed; workers earn livelihoods. The analogy is illuminating but incomplete.
Whether productivity creates employment depends on the elasticity of demand. If AI reduces the cost of a service but demand for that service is already saturated, output will not expand sufficiently to preserve employment. A company may need more AI-assisted cybersecurity analysis because threats are virtually unlimited, but it may not need ten times as many routine internal reports merely because reports have become cheaper.
Economists Daron Acemoglu and Pascual Restrepo offer a more nuanced account. Automation produces a displacement effect when machines take over existing human tasks. Productivity gains may increase demand for labour elsewhere, while the creation of new human tasks generates a reinstatement effect. Employment outcomes depend on the balance among these forces—not on automation alone.
Joseph Schumpeter’s idea of creative destruction offers a broader historical perspective. Innovation destroys established firms and occupations while creating industries around new products. Yet creation and destruction are rarely simultaneous or geographically balanced. The typist displaced in one city does not automatically become a machine-learning evaluator in another.
The “compensation theory” of technological change, advanced by Marx and later established by David Ricardo, similarly argues that employment losses are eventually offset by lower prices, higher investment, rising incomes, and the manufacture of new goods. Its weakness lies in the word “eventually.” A society can experience long periods of unemployment, wage pressure and political instability before compensation occurs.
Baumol’s cost-disease theory offers another counterpoint. As AI makes some sectors highly productive, labour-intensive human services may become relatively more expensive. This can increase wages in specialised care, education and performance—or make these services inaccessible unless governments and households are willing to pay for them.
The future is therefore unlikely to be either mass technological unemployment or effortless abundance. It will be characterised by simultaneous displacement, expansion and occupational mutation.
Layoffs and the Destruction of Career Ladders
The immediate corporate response has often been less subtle. Companies are reducing staff, consolidating functions, and redesigning roles to support AI-driven workflows. Oracle, for example, reported a reduction of roughly 21,000 employees between May 2025 and May 2026, while expanding investment in AI infrastructure. Salesforce’s leadership said its customer-support workforce had fallen from about 9,000 to 5,000 as AI agents handled a substantial share of interactions.
Not every layoff attributed to AI is caused by technical substitution. Some reflect post-pandemic overhiring, slowing demand, shareholder pressure, mergers or the need to finance expensive data centres. “AI” can become a convenient legitimising narrative for conventional cost-cutting.
Nevertheless, the organisational consequences are real. Companies are not merely eliminating jobs; they are decomposing roles. A marketing employee may retain strategy but lose drafting and basic analysis. A programmer may supervise the generation of code rather than write every component. A manager may oversee a smaller human team and a larger set of automated agents.
The gravest danger is the erosion of entry-level work. Junior employees traditionally learn through research, documentation, first drafts, routine cases and supervised mistakes—the very tasks organisations are most eager to automate. If firms remove the bottom rungs of the career ladder, they may achieve short-term savings while destroying the process through which future experts are developedthe ability to draft and perform . An economy cannot indefinitely demand experienced professionals while eliminating the opportunities through which experience is acquired.
From Digital Divide to Knowledge Differential
A second fault line is emerging between countries. AI advantage depends not only on internet access but on computing infrastructure, reliable electricity, data, language coverage, payment systems, institutional capacity and access to frontier models. Anthropic’s Economic Index has found that Claude usage is disproportionately concentrated in high-income countries. This suggests that AI may initially widen, rather than narrow, international productivity gaps.
Reports in June 2026 described Anthropic’s Mythos 5 as a restricted-access frontier system and Fable 5 as a safeguarded version, with their availability tied to US security and export-control concerns. Because official public documentation remains limited, claims about their capabilities and restrictions should be treated cautiously. The broader issue, however, is unmistakable: frontier intelligence is becoming a geopolitically controlled resource.
If advanced models are available mainly to approved organisations or selected nationalities, workers in privileged jurisdictions will learn to collaborate with more capable systems. They will generate better research, software, patents, and educational materials, accumulating experience that cannot be replicated merely by gaining access several years later.
This creates a knowledge differential, deeper than the conventional digital divide. The digital divide concerns access to devices and connectivity. The knowledge differential concerns the quality of cognitive tools available, the ability to use them, and the institutional learning accumulated around them.
Developing countries could thus face a double disadvantage. Their routine service exports may be automated by firms in richer economies, while their own entrepreneurs and universities remain unable to access the systems required to move into higher-value activities. Linguistic underrepresentation will compound the problem: AI trained primarily on dominant languages may be less capable of interpreting local laws, cultures and commercial contexts.
Yet an alternative is possible. Open models, regional computing facilities, multilingual public datasets, shared evaluation centres and national AI literacy programmes could enable developing economies to adapt AI to local needs. India and other populous developing nations possess an important asset: enormous diversity of languages, markets, social problems and entrepreneurial use cases. With suitable infrastructure, this complexity can become an innovation advantage rather than a handicap.
Governing the Employment Transition
The decisive policy question is not how to preserve every existing job. Many jobs contain exhausting, dangerous or intellectually sterile tasks that deserve to disappear. The objective should be to preserve human agency, mobility and dignity during the transition.
Governments and companies should therefore:
- measure automation at the task level rather than by job title;
- preserve apprenticeship pathways even when junior work can be automated;
- link productivity gains to wage growth, reduced working time, or profit-sharing;
- provide portable learning accounts and transition support before displacement occurs;
- treat frontier AI access as an element of development policy;
- invest in multilingual models and public-interest computing infrastructure;
- establish international rules that distinguish legitimate security restrictions from technological protectionism.
Jevons’ paradox offers conditional hope, not a guarantee. AI can expand demand for intelligence-based services and generate forms of work that are currently difficult to imagine. But markets do not automatically distribute new opportunities to the people, occupations or countries that bear the losses.
The future of employment will be determined not only by what intelligent machines can do, but by who owns them, who can access them, which tasks institutions choose to automate and how the gains are shared. AI may create an age of expanded human capability—or a world in which inexpensive machine intelligence coexists with diminished human opportunity. That choice remains profoundly political.