The Jevons Paradox: Why We Work More with AI

VZ Lens

Through a VZ lens, this analysis is not content volume - it is operating intelligence for leaders. AI promises to boost productivity. Research data shows that people who use AI more work more, not less. An extra 3 hours a day. The practical edge comes from turning this into repeatable decision rhythms.

TL;DR

According to a longitudinal study by the NBER, heavy AI users work 3.15 hours more per day and have 3.20 hours less free time. AI doesn’t free us up—it extends the boundaries of work. This is a modern version of the Jevons paradox.

---

Morning Crowd on the Ginza Line

My shoulder presses against the plastic window; the steam from my breath leaves a smudge on the glass. Outside, the Tokyo morning is gray and bright, the neon lights still flickering in the early morning mist. In front of me, I see the back of a businessman’s neck; a tiny lint speck sits on the collar of his white shirt. Next to him, a student’s finger flits across his phone, text scrolling down the screen—something artificial, fast, efficient. The subway’s rumbling is steady, but the tension in the air is palpable. We’re all rushing to get there on time, to get started. In my hand, the list of daily tasks flashes on my phone, which I put together last night with the help of AI—it’s shorter, but somehow more urgent. Lulled by the illusion of speed, rushing toward the goal, it occurs to me: where are we rushing to so much if we’re speeding everything up?

Dawn silence in the garden

I sit on the frosty wooden bench, a warm cup in my hand. The garden is still sleepy, dew glistens on the tree branches. The first rays of the sun slowly tear through the fog on the hillside. I hear the morning chirping of birds, the distant crowing of a rooster. The air is clear and crisp. I take deep breaths.

And yet—the day’s plan is already spinning in my head. Yesterday’s tasks, supplemented by today’s. Despite the silence, I cannot rest. Instead of nature’s rhythm, my internal clock is racing, like some wound-up mechanism. The sun is rising, but I’ve been at work for a long time.

Midnight, inbox empty

It’s midnight. My inbox is empty. Copilot helped me handle today’s thirty-two emails, ChatGPT summarized the four documents, and Claude wrote the draft proposal.

I’ve done it all. I was more efficient than ever.

And yet—I’m more tired than when these tools didn’t exist.

The paradox is simple: what AI saves me, I don’t use for rest. Instead, I use it for more work. Because if the proposal is ready in one hour instead of two, I don’t get an extra hour off—I get another task that I can now “fit in.”

William Stanley Jevons in 1865

Jevons, an English economist, observed in 1865 that improvements in the efficiency of the steam engine did not reduce coal consumption. It increased it. Because the more efficient steam engine made more uses economically viable, and the sum of all those uses exceeded the savings.

This was true for coal 160 years ago. Today, it is true for human attention.

AI does not reduce cognitive work. It makes cognitive work that we would not have done before economically viable. The analysis written by AI that we would have skipped before. The presentation created by AI that previously “didn’t fit.” The three proposal versions generated by AI, whereas before we would have created just one.

The economic root of the paradox: Why can’t we be forward-looking?

Jevons essentially argued that human decision-making is not based on grand strategic plans, but is driven by microeconomic incentives. When something becomes cheaper or easier to obtain, we consume more of it. The fact that coal became cheaper did not result in people doing more work and resting with the same amount of steam, but rather in new, previously uneconomical uses coming into practice. Exactly the same thing happens with our mental energy. A quote from the corpus [UNVERIFIED] perfectly illustrates this in another sphere: “Economic logic would dictate that taxi drivers work hard on rainy days and take it easy on sunnier days… However, the logic of loss aversion leads to exactly the opposite.” Taxi drivers work longer on bad days to avoid losses, while missing out on the profits of good days. It’s the same with AI: we don’t “capitalize on” the time saved, but instead invest it in even more work due to loss aversion (“what if I don’t do this too?”).

The Numbers: Not Opinions, but Time-Use Data

The NBER study is not based on opinions—it uses longitudinal time-use data.

Those who use AI intensively: +3.15 hours of work per day. The same people: -3.20 hours of free time per day. The efficiency gains do not translate into rest—they translate into work.

These numbers are not just statistics; they reflect a cultural and organizational response. A quote from the corpus [CORPUS] supports this: “After 1855, the data is for the UK… the full-time employment of women has increased over the past several decades. How did technology mysteriously descend on us?… all of us contribute to the chaotic system and impose more work on ourselves.” Technological progress has, in fact, never automatically led to shorter working hours. Since the 1980s, even amid the digital revolution, working hours have either increased or stagnated in the developed world. AI is merely the latest chapter in this long-standing trend.

Reddit communities reflect this exactly. One software developer wrote: “Copilot hasn’t freed me up. I’m just doing the work of two people for one person’s salary now.” Another: “AI is the perfect tool for the productivity trap.”

Why don’t you feel the expansion? The invisible link

The AI version of the Jevons paradox has a sneaky feature: you don’t feel the expansion. Every single micro-decision—“I’ll do this too, because the AI does it quickly”—is rational in itself. The accumulation is invisible.

No one decided to work three hours more per day. Every ten minutes saved led to another ten-minute task. The steam engine didn’t tell the coal it had to burn more. The coal simply burned because the machine made it possible.

The Curse of Work Flexibility

Cognitive work is fundamentally flexible. It has no physical limits, unlike an assembly line. AI takes this flexibility to the extreme. Previously, the amount of work we could do was limited by the 24 hours in a day and the finite capacity of our brains. Today, AI expands this capacity, but expectations and our own conscience immediately fill the new space. Another [UNVERIFIED] section of the corpus highlights this: “With that in mind, if you had an additional hour in your day, how would you spend it?… Would you brush up on professional development?… refine some of your processes?… Maybe you’d make personal phone calls…?” The question itself is misleading: it assumes that the time saved will be spent on work. Because organizational culture—and often our own ambition—suggests that “free” time must still be spent productively.

How does this expansion work at the corporate level?

The paradox isn’t just the result of individual decisions. It’s reinforced by corporate decision-making and the “machine time” paradigm.

A [CORPUS] quote describes how AI at AT&T finds “lost hours”: “At AT&T, AI is finding lost hours in workers’ days by pulling sales leads across multiple systems. This has freed up sales associates to spend more time developing relationships with customers…” This is an ideal scenario where the time saved is spent on quality work (building relationships). But the text immediately adds the critical caveat: “But if managers and CEOs continue to operate in the paradigm of machine time, then workloads will likely increase.” Here’s the crux: the “machine time” paradigm means that if a task is 30% faster, the expectation is for 30% more output, not 30% more thinking time or quality interaction.

Companies often do not reduce headcount but instead expand job responsibilities. Another excerpt from the corpus [CORPUS] confirms this: “we did not encounter any indication from any of these companies that they were reducing their overall headcount. In fact… a number of the companies gradually increased their overall headcount as a result of growing business opportunities”. AI, therefore, does not necessarily eliminate jobs, but rather enables growth. The problem is that this growth often means only more tasks and expanded responsibilities for the individual employee, without a proportional increase in compensation or time off.

What do you do when efficiency becomes a trap? The art of the deliberate limit

Reducing efficiency does not help against the Jevons paradox. Rather, it is the deliberate limit.

The question isn’t how much you could do with AI. It’s how much you want to do. If you don’t decide in advance, AI will decide for you—by making everything possible.

Practical steps to break the invisible chain

1. Setting time quotas: Decide how much time you will allocate to a specific type of task (e.g., email, report writing). If the AI cuts the time in half, do not spend the remaining time on additional tasks of the same type. Mark this physically in your calendar: “saved analysis time - strategic thinking”.
2. Introducing quality thresholds: The [CORPUS] quote highlights that AI-supported employees value work that is “more independent” and “more intellectually stimulating”. Use AI not to increase quantity, but to enhance quality. For example: use ChatGPT not to write 5 template emails, but to write 1 email with 3 different narratives and select the best one.
3. Redefining personal ROI: Ask yourself: If AI saves me 10 hours a week, where would I invest those 10 hours to maximize my long-term value (professional and personal)? The answer could be: learning, networking, creative prototyping, or simply relaxing. Treat this investment as just as important as your work tasks.
4. Initiate a company-wide dialogue: Talk to your manager or team about the “machine time” paradigm. The question is: should we use the freed-up capacity to increase quantity (more output) or quality (smarter output, innovation, employee well-being)? The corpus citation [CORPUS] refers to the “Productivity J-Curve”, which suggests that the full impact of breakthrough technologies (such as AI) is realized only with a delay, through new organizational forms and skills. This requires time and conscious planning.

Beyond Productivity: What Comes Next?

The lesson of the Jevons paradox is that technological progress alone does not lead to a happier or freer society. Freedom is not granted by technology, but by our personal and collective decisions. The factory laws of the 19th century and the introduction of the 8-hour workday also required technological progress, but their implementation demanded political will and social movements.

In the age of AI, we will also need a new “social contract” regarding work and time. This does not necessarily mean a universal basic income, but rather a redefinition of expectations, compensation, and leisure time. As CORPUS notes: “Everyone talks about the world of work as being threatened by robotics and automation. But… long hours made a comeback in the 1980s as a kind of macho, neoliberal thing.” AI offers an opportunity to break free from this “macho” concept of productivity, but only if we consciously choose the other path.

Key Takeaways

• AI users work +3.15 hours more per day—not less. This is the cognitive version of the Jevons paradox.
• The Jevons paradox: increased efficiency does not reduce but rather increases total consumption, because it enables new ways of using resources.
• The expansion is invisible: we end up with more working hours as a result of a series of micro-decisions, not a single major decision. The “machine time” paradigm reinforces this in companies.
• The antidote is not less AI, but setting deliberate limits: deciding how much you want to do, not how much you can. The time saved must be consciously invested in quality work or recovery.
• The long-term solution requires not only individual but also organizational and societal dialogue on how we define productivity and well-being in the post-AI era.

Frequently Asked Questions

What is the Jevons paradox, and how does it affect AI?

In 1865, William Stanley Jevons observed that while the efficiency of steam engines had increased, so had coal consumption. Why? Because the more efficient machines were used in more places, which more than offset the savings per unit of work. The same thing happens with AI: we fill the cognitive time saved with new cognitive tasks, which leads to a net increase in work. According to NBER data, AI users work 3.15 hours more per day.

Why do we work more with AI if it’s more efficient?

This works on two levels. Individually: our conscience or ambition compels us to use the saved microseconds for new tasks (“we can fit it in now”). Organizationally: the “machine time” paradigm leads management to translate increased efficiency into expectations for increased output. The expansion is invisible: you write an email faster, so you write another one. You prepare a report faster, so you prepare another one.

Does this mean AI is bad?

Not at all. AI is an extraordinary tool. The issue isn’t the tool itself, but its use. AI is neutral in and of itself. The paradox isn’t AI’s fault, but a natural consequence of human and organizational decision-making. The challenge is to become aware of this pattern and intentionally shape our use of AI to promote freedom, not an even greater workload.

What is the antidote to the productivity paradox?

The key is to set intentional limits and repurpose the time saved.

1. At the individual level: Decide in advance how you will spend the time saved today (e.g., strategic thinking, learning). Set time limits for certain types of tasks.
2. At the team/organizational level: Initiate a dialogue about whether to use the increased capacity to improve the quality or quantity of work. Refer to the long-term “Productivity J-Curve” concept [CORPUS], which states that true efficiency gains require time and organizational transformation.
3. On a psychological level: Accept that the flexibility of work is infinite, but your capacity and values are not. Use AI to serve your own value-based priorities, not to cater to external (or internal) expectations.

Are there examples of AI actually providing free time?

Yes, but this is generally not an automatic outcome, but a planned one. In the corpus [CORPUS], the example of AT&T shows that AI identified “lost hours,” and management decided to use this time to build relationships with customers—which is more value-creating, but not necessarily “more” work. In successful cases, the organization consciously redesigns the workflow to automate routine tasks and focus human resources on more complex, creative, or social tasks.

---

Related Thoughts

• AI Burnout Protocol: 5 Practical Steps
• The Adoption-ROI Paradox
• Why Do 90% of AI Projects Fail?

---

Zoltán Varga - LinkedIn Neural • Knowledge Systems Architect | Enterprise RAG architect PKM • AI Ecosystems | Neural Awareness • Consciousness & Leadership The pattern survives. The noise compounds.

Strategic Synthesis

• Define one owner and one decision checkpoint for the next iteration.
• Track trust and quality signals weekly to validate whether the change is working.
• Run a short feedback cycle: measure, refine, and re-prioritize based on evidence.