AI, Jobs, and the Jevons Paradox: Why the Void Rarely Stays Empty

Why efficiency-driven job loss is only the first act

---

AI, Jobs, and the Jevons Paradox: Why the Void Rarely Stays Empty

Why efficiency-driven job loss is only the first act

I first heard Jevons Paradox at a group dinner from Greg Sands, founder and managing partner at Costanoa Ventures, and it hit me with that rare feeling economists don’t often deliver: oh — that explains it. Not in an abstract way, but in a pattern-recognition way. I’d seen this movie before. I’d watched supposedly job-destroying technologies do the opposite over and over again: electrification turning factories into engines of mass employment, the PC and spreadsheets “eliminating” accountants only to create entire finance and software industries, the internet hollowing out media jobs while spawning e-commerce, digital marketing, cloud computing, and the modern startup economy. Each time, a tectonic innovation made something radically cheaper, triggered panic about displacement, and then—almost embarrassingly—ended up raising productivity, expanding demand, and lifting far more boats than it sank. AI suddenly snapped into that same historical groove, and Jevons Paradox gave the pattern a name to me.

Jevons Paradox is the idea that when a technology makes something dramatically more efficient and cheaper, total usage of that thing often increases rather than decreases because lower costs unlock new demand and new uses.

Everyone has a take on AI and jobs right now, and most of them rhyme.
This time is different. AI will replace workers permanently. Productivity will rise, but employment won’t. The gains will flow to capital, and the labor market will hollow out.

It’s a clean story. It’s also missing a crucial piece of economic history.

There’s a 160-year-old idea that complicates the doom loop in an important way: Jevons Paradox. Take it seriously, and the AI future looks less like a permanent job apocalypse and more like something stranger, messier—and ultimately more abundant.

“I just want to say one word to you. Just one word.”
“Yes, sir?”
“Plastics.”

In The Graduate, plastics stood in for a future that felt inevitable, vaguely unsettling, and massively transformative. AI is our plastics moment. Not because it replaces everything, but because it ends up everywhere.

---

Jevons Paradox, Applied to Intelligence

Jevons Paradox observes that when a technology makes something cheaper and more efficient, total usage often increases rather than decreases. When steam engines became more efficient in the 19th century, Britain didn’t burn less coal. It burned far more, because energy-intensive activity exploded.

Now swap coal for cognition.

AI radically lowers the cost of thinking tasks: writing, analyzing, coding, summarizing, planning, deciding. The marginal cost of large chunks of white-collar work is collapsing toward zero.

In the short run, this absolutely destroys jobs. But Jevons tells us that’s only the first act.

---

How AI Really Cuts Roles

Let’s be clear-eyed. AI does substitute for labor.

Firms are already using AI systems to automate portions of customer support, compress layers of reporting and analysis, accelerate software development, and redesign jobs around supervision rather than execution. At the firm level, AI adoption often coincides with restructuring, hiring freezes, and selective job elimination—especially in roles composed of highly automatable tasks.

A customer-support team that once needed 100 agents may now handle the same ticket volume with 70, simply because each remaining worker is dramatically more productive. Similar patterns are showing up across marketing, finance, legal operations, and engineering.

This is the part everyone sees. And it’s real.

---

Why the “Void” Doesn’t Stay Empty

Here’s where Jevons Paradox reenters the picture.

When the cost of cognition falls, demand for cognition doesn’t politely hold steady. It expands. Lower thinking costs mean cheaper software, more customization, more experimentation, more personalization, and faster iteration. And when those things get cheaper, organizations don’t stop—they do more.

“When the cost of thinking collapses, perhaps demand for thinking doesn’t disappear — it explodes.”

AI-using firms tend to grow faster, gain market share, and attempt projects that were previously uneconomical. Entire categories of products and services suddenly make sense. Internal constraints shift from “we can’t afford to think about this” to “we can try ten versions by Friday.”

Empirically, firms that use AI most intensively are often larger, more productive, and—over time—grow headcount faster than peers, even as specific job families shrink or mutate.

This is Jevons in labor-market form: efficiency in one task expands the frontier of what the organization can attempt. That expansion absorbs much of the “freed” capacity, but in new roles, new workflows, and new sectors.

---

Why This Still Doesn’t Save Every Job

The paradox is powerful, but it’s not automatic or universal.

Where AI is a near-complete substitute, demand for that specific human role can remain permanently depressed. In other areas, physical constraints—energy, materials, regulation, time—limit how much activity can scale, even if cognition becomes cheap.

Timing and distribution matter too. Job losses tend to be sharp and localized, while Jevons-style job creation unfolds slowly and unevenly, often in different regions and occupations. That gap is where real pain lives.

So yes, we can have both: permanent losses in certain roles and aggregate employment growth as AI-enabled sectors expand.

---

The Demographic Twist Everyone Misses

Here’s the underappreciated irony.

The United States isn’t heading toward a future of surplus labor. It’s facing slower population growth, rising dependency ratios, and potential worker shortages. Fewer workers, more retirees, and sustained demand for services.

In that context, AI looks less like a job-destroying catastrophe and more like a stabilizing force.

“Perhaps AI will destroy jobs in the short run, then quietly expands the frontier of what’s worth doing.”

AI-driven productivity allows a smaller workforce to support a larger economy. It offsets demographic drag without collapsing living standards. It makes growth possible even when labor supply tightens.

Too little productivity growth in a shrinking workforce leads to stagnation. Too much automation in a rapidly growing population leads to unrest. AI, arriving just as labor growth slows, sets up a strange Goldilocks scenario: enough automation to maintain growth, but enough unmet demand and new use cases to keep humans economically essential.

---

The Question That Actually Matters

The real question isn’t whether AI will kill jobs.

It’s where AI acts as a substitute, and where it acts as a complement that expands demand. In the complementary zones, Jevons-style dynamics dominate. Efficiency doesn’t reduce human relevance—it amplifies it.

Humans become problem definers, system orchestrators, trust builders, risk managers, and interpreters between increasingly complex systems and real-world needs.

The paradox of AI and employment is this: the very efficiencies that eliminate some jobs create the economic space—and often the necessity—for new ones to grow into the void they leave behind.

“The void left by efficiency rarely stays empty; it fills with new roles, new work, and new ambition.”

Plastics didn’t end manufacturing. Cheap computation won’t end work. It just changes what’s worth doing. And historically, when we make something vastly cheaper, we rarely decide to want less of it.

“Every major technology panic mistakes substitution for extinction — and AI is no different.”

I don’t pretend this transition will be smooth, fair, or painless for everyone. History rarely is. But if Jevons Paradox is any guide, the void left by efficiency doesn’t stay empty for long. Cheaper cognition doesn’t shrink the frontier of human work—it pushes it outward in unpredictable directions. I’m curious, cautiously optimistic, and very much glass-half-full about where that leads. Over the years to come, we’ll see whether AI follows the same messy but ultimately expansive path as past general-purpose technologies. My bet is that it does—and I hope I’m right.