When James Watt introduced a much more steam-efficient engine, everyone thought that since the engine needed less coal, total coal consumption would drop. William Stanley Jevons—an English economist—absurdly noted, however, that because using more efficient engines made using coal cheaper, factories found more ways to utilise the resource, causing total coal consumption to skyrocket rather than drop.
In the context of Artificial Intelligence, the resource is not coal; it is human time, creativity, and cognitive labour. Before AI, executing a task required a high cost of human hours. AI drastically lowers that cost; the shift is almost dramatic. True to the Jevons Paradox, this efficiency has not resulted in humans doing less and creating the same amount of output. Instead, because the cost of creating output has plummeted, the demand for output is at an all-time high, and it continues to rise.
The collapse of the marginal cost.
In economics, the "marginal cost" is the cost of producing one additional unit of a good. AI is effectively driving the marginal cost of standard cognitive tasks such as writing code, drafting copy, analysing data sheets, and creating visual templates, very close to zero. When the cost of generating an asset collapses, organisational behaviour shifts from preservation to experimentation. A company might have had the budget to write and test two marketing campaigns per quarter.
With AI lowering the friction, the same team is tasked with deploying two hundred hyper-personalised variations targeted at micro-niches simultaneously. The worker's time is not saved; rather, their productive frontier is aggressively pushed outward. But this expansion goes far beyond simply producing more assets. The true paradox of AI efficiency lies in how it fundamentally alters, intensifies, and complicates the architecture of human labour.
The trap of pseudo-work and cognitive triage.
The most insidious side effect of near-zero marginal costs is the generation of pseudo-work, outputs that perfectly mimic highly productive activity and appear completely free within digital networks, yet are strategically irrelevant. Because AI makes generation effortless, we are drowning in complex, superficially plausible data sheets, software code, and texts that arguably never needed to exist in the first place. AI sustains an illusion of value, but it shifts the human burden from creation to what is now known as cognitive triage.
Humans have become the filters for automated clutter. Instead of spending hours writing, professionals now spend hours sorting, filtering, validating, and performing multi-agent verification workflows just to find the rare kernel of strategic truth amidst a flood of synthetic noise.
The trust crisis and the jagged frontier.
This verification workload is amplified by what Demis Hassabis describes as the "Jagged Frontier" of AI intelligence, the phenomenon where an AI can flawlessly execute a highly complex, advanced task but catastrophically fail at a simple, basic one. Because of this jaggedness, AI disrupts traditional trust loops. In fields like software engineering, law, and academic research, AI can instantly generate hundreds of plausible proofs or arguments that are formally consistent but humanly incomprehensible. This creates a verification burden: Because a machine can hallucinatory generate errors that look identical to brilliant insights, human teams must deploy sophisticated, exhausting error-correction loops. Reviewing, debugging, and de-risking automated output is often far more mentally taxing than executing the work manually from scratch.
The superstar paradox and the commodity of competence.
AI has democratized access to powerful execution tools, making it easier than ever to attempt to be a high-performing creator, programmer, or researcher. However, this creates what economists call the Superstar Paradox: as competence becomes commoditised at scale, actual visibility shrinks. If AI tools increase the total volume of plausible, high-quality output in a field by 10 times, they flood the ecosystem. In reality, while the supply of excellent work increases tenfold, humanity's collective attention budget remains completely fixed. Consequently, the probability of any single excellent piece of work being seen or read plummets. In an AI-augmented landscape, baseline excellence is no longer sufficient. To break through the high-quality noise, human output must deviate into the transcendent, demanding even higher levels of human taste, risk mitigation, and profound creative genius to stand out.
The dissolution of natural pauses.
On a day-to-day level, AI efficiency does not grant leisure; it intensifies the cadence of the workday by eradicating natural cognitive boundaries. As researchers observe, when a tool makes it frictionless to start and continue tasks, work naturally bleeds into moments that used to function as pauses. By keeping multiple threads alive simultaneously, where the human is continuously prompting, editing, and monitoring while the machine processes in the background, both human and machine are kept in a state of constant, exhausting motion.
The naive view of automation assumes that every task taken over by AI is a net substitute for human effort, leading straight to a lighter workload. But this completely ignores composition effects and negative feedback loops.
AI efficiency does not design a world with less labour. By minimising the hours spent on raw generation, it expands the volume of clutter we must filter, the boundaries of the roles we must fill, and the threshold of quality required to be seen. We are not working less; we have simply been transformed from computational labourers into deep strategic conductors, trapped in a vacuum where efficiency breeds abundance, and abundance demands infinite human curation.
References
- Jevons, W. S. (1865). The Coal Question: An Inquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines. London: Macmillan and Co.
- Owen, D. (2010). "The Efficiency Dilemma: If our machines use less energy, will we just use them more?" The New Yorker (December 20/27 Issue).
- Hidalgo-Pérez, M. A. (2026). "The AI-Driven Skill Premium: A Model of Positional Scarcity and Cognitive Traps," Working Papers 26.01, Universidad Pablo de Olavide, Department of Economics.
- Ye, X. M. (2026). "The multi-threaded workday: Qualitative insights into generative AI workplace intensification" (Field Brief). UC Berkeley Haas School of Business.
- Zhang, Y., & Zhang, T. (2026). "The Economics of Digital Intelligence Capital: Endogenous Depreciation and the Structural Jevons Paradox," arXiv Preprint arXiv:2601.12339.
- Fiehler, J. (2026). "Jevons paradox: Rise of the machines may create more work for physicians." Journal of NeuroInterventional Surgery, 18(2).
- Khosravirad, S. R. (2026). "Researcher in the age of AI: From computational labourer to conceptual orchestrator" (Preprint). Preprints.org.
- Mann, H. (2026). "AI productivity blind spot: Composition effects, cognitive externalities, and negative feedback loops." California Management Review Insights. University of California, Berkeley.
- Muhlert, M. (2025). "The convergent singularities: Diagnosing the meta-crisis, heteromation, and pseudo-work." Advances in Corporate Governance, 12(4), 88–97.