Jevons Paradox and AI, Explained Simply for Business Owners

TypeSafe AI named its new decision model Jev, after a 160-year-old idea about coal. On a recent a16z podcast, founder Diogo Almeida explained the name: once routine work becomes automatable, “people will just make… new kinds of work, hence the Jev in Jevons.”

That idea, Jevons paradox, is now one of the most-quoted arguments in AI. It’s also one of the most useful for a business owner, because it explains why cheaper AI won’t shrink your AI use. It will grow it, and the shops that see that coming get the head start.

What is Jevons paradox, explained simply?

In 1865, a British economist named William Stanley Jevons published The Coal Question. Britain was worried about running out of coal, and the popular hope was that more efficient engines would save it. Jevons said the opposite would happen:

“It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth.”

His best example will sound familiar to anyone in steel. Scottish ironworks adopted the hot blast, which cut the coal needed per ton of iron to less than a third. Did Scotland burn less coal? No. Jevons wrote that it was “followed, in Scotland, by a ten-fold total consumption.”

He spelled out the chain reaction too: when a blast furnace needs less coal for the same yield, profits rise, new capital comes in, the price of pig iron falls, demand grows, and “the greater number of furnaces will more than make up for the diminished consumption of each.”

Think of it like getting a faster beam line. You don’t cut the same tonnage in fewer hours and send everyone home. You bid more work, because now you can.

Why is everyone in AI talking about Jevons paradox?

Because AI is getting cheaper faster than almost anything in history, and usage keeps climbing anyway.

• The price is collapsing. Epoch AI’s September report, The plunging price of thought, estimates the cost of a given level of AI performance has fallen ~47% per quarter since 2023, or ~13x a year. Their comparison: like a new car’s sticker price dropping from $50,000 to $69. (We dug into what that means for growth in Dream Bigger.)
• Use is exploding. At Google I/O in May, Sundar Pichai said Google was processing over 3.2 quadrillion tokens a month, up from roughly 480 trillion a year earlier and 9.7 trillion two years before that. (A token is a small chunk of text, often part of a word.)
• The big players are betting on it. When DeepSeek’s cheap models rattled markets in January 2025, Microsoft CEO Satya Nadella posted: “Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket.” (More on DeepSeek and cheap open models in our open vs closed models post.)

Epoch even notes the flip side: the AI boom is pushing up prices for what it consumes, including chips, power and “even the labor of electricians.” Cheaper AI, more data centres, more demand. Coal all over again.

What does Jevons paradox mean for a fab shop?

Here’s the practical version: cheap AI makes it worth doing the work you skip today.

Every shop has a list of things that would be nice but aren’t worth a person’s hour. Cheap AI changes the math on that list:

• Bidding. Today your estimator passes on some bid invites in a busy week. When an AI agent can draft a takeoff and summarize the scope, you can look at every invite and bid the ones that fit.
• Mill certs. Today someone spot-checks certs against the grade ordered. When checking costs pennies, you check every heat number on every cert.
• Drawing revisions. Today someone eyeballs the new set. When comparison is cheap, every revision gets a piece-by-piece diff with tonnage impact.
• Customer updates. Today the GC calls asking “where’s my steel?” When drafting status reports is cheap, every GC gets one every week, before they ask.

Notice what happened. The cost per task fell, but you’re running more tasks than before. That’s Jevons in your own office. (Our guide to what to automate around Tekla PowerFab goes workflow by workflow.)

Almeida’s own pitch for Jev is blunt: “where the f*** is all the automation?” He told a16z that “intelligence per dollar” is his north star right now. That’s a bet on Jevons: make each decision cheap enough and people will automate decisions nobody bothered with before.

Will AI reduce jobs, or shift them?

This is the question people search most, and history gives an interesting answer. Jevons himself made the point in the same chapter: new machinery “throws labourers out of employment for the moment,” but demand for the cheaper products grows so much that “the sphere of employment is greatly widened.”

Three examples:

• Power looms. Economist James Bessen notes that in the 19th century, power looms automated 98% of the labour needed to weave a yard of cloth. Yet factory weaving jobs increased, because cheaper cloth meant far more demand for cloth.
• ATMs. ATMs cut the tellers needed per urban branch from about 20 to 13 between 1988 and 2004. But branches got cheaper to run, so banks opened more: urban branches rose 43%, and teller jobs didn’t disappear. Tellers shifted toward customer relationships.
• Radiology. In 2016, AI pioneer Geoffrey Hinton said “people should stop training radiologists now.” Instead, the New York Times reported that Mayo Clinic’s radiology staff grew 55% to about 400, while the clinic uses AI to work faster.

The pattern: the job changes shape. a16z’s Alex Danco put it this way: some parts of a job get automated and see 10x the throughput, while the part only a human can do “will be the reason you’re getting paid.” In a fab shop, that’s your estimator’s judgment on a tricky connection, your PM’s relationship with the GC and your QA lead’s sign-off.

Does Jevons paradox always happen?

No, and it’s worth being honest about that.

It needs demand that can grow. Jevons works when cheaper output unlocks a lot more wanted output. Cheaper bids can win you more jobs. But cheaper invoice processing won’t make you send 10x more invoices; that saving simply drops to the bottom line. Both are fine. Just know which is which.

It’s good news for buyers, not a promise to sellers. Investor Paul Kedrosky, in a Big Technology Podcast clip written up by FourWeekMBA last month, argued that saying “but Jevons’ paradox, but people will use more” is a way “to really dodge the core problem of the geometric decline in the price.” That’s a worry for whoever sells AI. For a shop that buys it, falling prices are the whole point.

More decisions means more total mistakes, unless you check. If AI handles 10x the tasks at the same error rate, you get 10x the errors. That’s why we keep a person approving anything that touches a customer, a PO or your system of record.

Why do early adopters win under Jevons paradox?

Jevons saw this coming too. He warned that efficient inventions are “as open to our commercial competitors as to ourselves,” and are “more readily adopted by versatile foreigners than by English manufacturers bound by custom and routine.”

Swap “foreigners” for “the shop down the road” and it reads like 2026. Cheap AI is available to everyone at the same price. The advantage isn’t the tool. It’s being the shop that redesigns its work first: quoting more, answering faster and taking on jobs others turn down. Jevons quoted an old maxim of trade: “a low rate of profits, with the multiplied business it begets, is more profitable than a small business at a high rate of profit.”

How do you put Jevons paradox to work in your shop?

A simple first pass you can do this week:

1. Write your “not worth an hour” list. Bids you pass on, certs you spot-check, revisions you eyeball, follow-ups you never send, reports nobody has time to build.
2. Circle the ones where more volume means more revenue. Those are your growth plays. The rest are cost savings.
3. Pick one that’s high-volume and low-risk. Start there, with a person approving the output.
4. Budget for volume, not unit price. Cost per task will fall; your total use will rise. Ask any vendor: “If our usage goes up 10x, what happens to our bill, and how do we catch errors at that volume?”
5. Decide where freed-up time goes. More bids, more customer time, a new product line. If you don’t decide, the time just leaks away.

Want help finding your Jevons opportunity?

Embedding AI builds workflow automation with AI agents (new to the term? start with what an AI agent is) and purpose-built software for mid-market fabricators and manufacturers.

Book a free 30-minute consultation and we’ll go through your “not worth an hour” list together and pick the first workflow. Prefer email? Write to info@embeddingai.ca.

Sources

• William Stanley Jevons, The Coal Question, chapter VII, “Of the Economy of Fuel” (1866 2nd edition, Online Library of Liberty; 1865 excerpt, Yale).
• The a16z Show, AI Can Write Code. Why Isn’t Software Better? with Diogo Almeida, Ben Horowitz and Martin Casado (September 28, 2026); quotes from the auto-generated transcript published by 1% Better.
• Epoch AI, The plunging price of thought (September 22, 2026).
• Google, Sundar Pichai’s Google I/O 2026 keynote (May 2026).
• GeekWire, Microsoft CEO says AI use will ‘skyrocket’ with more efficiency amid craze over DeepSeek (January 27, 2025).
• James Bessen, Toil and Technology, IMF Finance & Development (March 2015).
• Radiology Business, NY Times revisits Nobel Prize winner’s prediction AI will render radiologists obsolete (May 15, 2025).
• Alex Danco, a16z, Why AC is cheap, but AC repair is a luxury (November 3, 2025).
• FourWeekMBA, The 100-Million-Fold Problem: Why “Jevons’ Paradox” Won’t Save the AI Chip Trade (September 15, 2026).

Frequently asked questions

What is Jevons paradox in simple terms?

When something becomes cheaper or more efficient to use, people find so many new uses for it that total consumption goes up, not down. Jevons noticed that better steam engines and blast furnaces used less coal per job, yet Britain burned far more coal overall.

How does Jevons paradox apply to AI?

AI is getting cheaper extremely fast. Epoch AI estimates the cost of a given level of AI performance has fallen about 13x a year since 2023. As each task gets cheaper, businesses use AI for many more tasks, so total AI use keeps climbing. Google reports processing over 3.2 quadrillion tokens a month, up from roughly 480 trillion a year earlier.

Does Jevons paradox mean AI won't take jobs?

Not exactly. It means cheaper work tends to create more demand, which has often shifted jobs rather than erased them. Power looms, ATMs and AI in radiology all arrived without shrinking those fields: weaving jobs and radiologist numbers grew, and teller jobs held steady. It isn't guaranteed, though. It depends on whether customers want much more of what you make once it gets cheaper.

What should a manufacturer do about Jevons paradox?

List the work you skip today because it isn't worth a person's hour, like jobs you don't bid, certs you only spot-check and follow-ups you never send. Those are the tasks cheap AI makes worth doing. Start with one high-volume, low-risk workflow, keep a person approving the output, and plan where the freed-up time goes.