In the 19th century, English economist William Stanley Jevons observed that technological improvements making coal usage more efficient actually increased overall demand for coal across industries. The paradox lies in the fact that, if demand were fixed, greater efficiency should reduce resource consumption. Yet the opposite occurred: efficiency unlocked new applications and use cases, driving massive growth. This pattern has repeated throughout history whenever we’ve made industrial processes, products, or technologies more productive and affordable. Especially in computing itself.
For example, in the mainframe era, only a few hundred units existed, affordable solely to the world’s largest corporations. With the arrival of minicomputers, smaller and less expensive, sales reached tens of thousands. By the early personal computer age, units sold in the millions. That’s roughly a 100x increase with each new computing era over just three decades.
In the 1970s, only Fortune 500 companies could access powerful accounting software. By the 2000s, cloud computing made it available to every small business worldwide, from local barbershops onward. The same democratization happened with CRM systems, communication tools, marketing automation, document management, and virtually every category of enterprise software. Overnight, the advantages large enterprises enjoyed in procurement, installation, maintenance, and computing power largely disappeared.
These efficiencies democratized the automation of deterministic work through software for decades across nearly every industry. However, non-deterministic knowledge work, the majority of daily enterprise tasks, such as reviewing contracts, writing code, creating advertising campaigns, conducting advanced market research, or providing 24/7 customer support, remained out of reach for widespread automation until now.
AI agents are bringing that same democratization to all forms of non-deterministic knowledge work, fundamentally transforming business. Today, large companies can easily reallocate resources across projects, experiment with new ideas, hire top lawyers or marketers as needed, and bring in engineers for any initiative. These capabilities, built over decades (or centuries) of success, have long given giants a structural advantage that puts most companies and entrepreneurs at a severe disadvantage from day one.
AI agents upend this equation by dramatically lowering the investment cost for nearly any organizational task. The common error in evaluating ROI is focusing on maximizing the “R” (return) when the true leverage lies in slashing the “I” (investment). Anyone who has managed budgets in a small team knows the painful trade-offs: choosing between a better marketing site, new product features, customer support, finance priorities, or distribution efforts. Each decision limiting growth in other areas due to scarce resources.
Now, AI removes the core constraint behind many of these trade-offs: the high cost of execution. Today’s consumers have better access to education and tutoring than historical aristocrats, thanks to AI. Similarly, every business worldwide now gains access to capabilities rivaling those of a Fortune 500 company from a decade ago.
Demand for many types of work will surge 10X or 100X as barriers fall, enabling companies to pursue projects previously deemed impractical. Consider a 10-person services firm that never built custom software due to the effort required in terms of e.g. development, maintenance, updates, security, and incorporation of feedback. Such projects often never started. Today, a team member can prototype one in days and validate its value almost immediately. The same logic applies across countless tasks and functions.
Many understandably worry about job losses in this shift. In reality, even as AI automates specific tasks, delivering true value still demands human orchestration of full workflows. AI agents require management, oversight, and rich context to realize their potential. Recent leaps in model performance have improved output quality, but fully autonomous AI capable of perfectly implementing and maintaining complex goals remains distant.
AI is already handling discrete tasks, like market research, coding features, or creating campaign media, but integrating them into broader workflows to generate real outcomes still relies heavily on human judgment and effort. As AI takes on larger portions of workflows, expectations for output will simply rise, turning today’s jobs into tomorrow’s routine tasks.
This dynamic is historically common. Tools like Figma or Google Ads would have seemed poised to eliminate marketing roles in the 1970s by consolidating multiple jobs into one. Instead, marketing employment exploded. Rough estimates (AI-assisted) show a few hundred thousand U.S. workers in marketing-related fields (PR, graphics, advertising, etc.) in the 1970s versus low millions today. This is a 5x+ increase over 50 years, precisely because technology made the work far more efficient.
Advertising shifted from an elite activity reserved for giant Consumer Packaged Goods (e.g. Nestlé, Procter & Gamble (P&G), Unilever, Coca-Cola, and PepsiCo), or automotive brands to something accessible to nearly any small business. Advances in marketing tech, CRM, analytics, design software, targeting, and distribution channels enabled far more companies to justify sophisticated efforts. AI will drive a similar expansion across many fields.
Jevons paradox is arriving in knowledge work.
By making virtually any imaginable task far cheaper, we’ll end up doing vastly more of it.
Jevons paradox is arriving in knowledge work. By making virtually any imaginable task far cheaper, we’ll end up doing vastly more of it. Most future AI tokens will power endeavors we don’t pursue today: software projects never launched, contracts never thoroughly reviewed, medical breakthroughs never explored, and marketing campaigns never attempted. Precisely because the efficiencies make them viable.
Leave a Reply