← narwal.one/Second Brain
SecondBrain
Ask the Brain
Index/Conceptupdated Tue Nov 03 2026 08:00:00 GMT+0800 (Philippine Standard Time)

Jevons Paradox

economicsgeneral-purpose-technologylabor-demandai-and-jobscoalelasticity

Jevons Paradox

William Stanley Jevons, 1865: as steam engines got more efficient, the demand for coal went up, not down. Falling cost per unit of work made steam power viable in more places; total consumption rose.

The general form: when a technology makes a complementary input cheaper, total demand for that input can rise if the applications of the technology are elastic enough. Falling cost per unit doesn't mean falling total quantity — it can mean the opposite.

Applied to AI (the 2026 framing this vault tracks)

Two vault sources use Jevons as the load-bearing argument that AI adoption expands rather than contracts labor demand:

Crume (IBM Technology) — the whiteboard version

AI Gave You A Promotion (IBM Technology): as AI makes each worker more efficient, cost of doing the task falls, but demand for the work people can now do rises — smart organizations reach more customers, ship more products, attempt more innovation. IBM's specific commitment (as reference for scale): "IBM is planning to hire two to three times as many entry-level people in 2026 as it did in the previous year, and that's with AI."

Crume's memorable line: "You're the coal." In the Jevons analogy, the human worker is the input whose falling per-unit cost drives up total demand.

Whittemore / Levie (Ramp/Revelio empirical anchor)

AI Companies Are Hiring More (AI Daily Brief) — AI Adoption and Headcount Growth is the concept — is the empirical version of the same argument. The Box CEO Aaron Levie mechanism, quoted on that page:

"AI wins more customers → hire more salespeople; ships more software → hire more engineers."

This is Jevons applied to labor: AI makes each salesperson/engineer more effective, so the marginal salesperson/engineer becomes profitable in places they weren't before, so headcount rises — for adopters.

Where Jevons might not apply (honest limits)

The whole argument turns on the elasticity of the applications:

  • Coal held Jevons. Steam-power applications in the 1800s were nearly unlimited; every new mill / mine / ship / rail was another consumer.
  • Some knowledge-work domains may be inelastic. If a firm has a fixed volume of legal review, and AI makes each unit of review 3× faster, and there is no market pull to expand what gets reviewed, then headcount falls. Jevons breaks.
  • Cross-firm reallocation. Even if AI expands total labor demand at the market level, individual firms may not participate — Ramp/Revelio's finding is that adopters hire, not that all firms hire. Non-adopters were flat, not shrinking, but the wedge is real. See Advantage Gap.

The safe operational reading is not "AI creates jobs" but "AI reallocates and can expand demand for the work of people who use it well; the risk is not headcount, it's whether your firm participates."

Related vault concepts

  • Six Elasticities of AI Demand — Whittemore's six-way breakdown of where the elasticity Jevons requires actually sits; a case-by-case test for the "inelastic domain" limit above.

  • AI Adoption and Headcount Growth — the empirical concept page; Jevons is the mechanism.

  • Code Is Free — the technology side. Implementation cost falls → implementation-supply elasticity rises → total implementation demand can rise.

  • Token Scarcity — a partial Jevons observation about tokens: falling per-token cost has driven up total token consumption (Ramp's 13× AI-spend growth). But rebounding-past-the-original-price — the pure Jevons form — is not universally observed yet; see the "tokconomics" chart.

  • FOBO (Fear of Becoming Obsolete) — the fear this argument speaks to. Jevons is the frame under which "AI won't take your job" is sometimes correct; it's not a blanket reassurance.

  • Solow Paradox — the sibling economic frame in this vault. Solow's J-curve is the why per-worker output doesn't rise; Jevons is the why total headcount can. Different quantities, both operational for the AI-and-labor question.

  • Deskilling Trap (Juniors) — the entry-level compression is real even if Jevons holds at the firm level: the internships / observe→assist→lead pathway (Standardized vs Open Tasks) is what preserves the pipeline.

Cross-links

Sources

  • AI Gave You A Promotion (IBM Technology) — Crume's whiteboard version + IBM's 2–3× hiring plan
  • AI Companies Are Hiring More (AI Daily Brief) — the Ramp/Revelio + Box CEO empirical anchor for the labor-side Jevons form