Jevons Paradox for Software

Jevons Paradox for Software

When AI drives the marginal cost of code toward zero, does total demand for software engineers contract or expand? The Jevons paradox is the strongest case for expansion, and it deserves to be taken seriously rather than waved away in either direction. My working position: bullish on total software, agnostic on any given role, and bearish on the old apprenticeship rung. That nuance is the honest place to stand.

The mechanism

The Jevons paradox, named for William Stanley Jevons in 1865 writing about coal, is the observation that efficiency improvements which lower the cost of using a resource can increase total consumption, because demand expands to more than absorb the per-unit saving. It applies when demand is elastic. Applied to software: if AI cuts the cost per feature, and appetite for software is nowhere near saturated, then total software built rises, and with it the work of framing, integrating, operating, and verifying it.

• Jevons' Paradox seems alive and well (Klement on Investing) is a clean explainer of the mechanism and its limits.

The bull case

Prominent voices in the industry, Microsoft's Satya Nadella among them, have argued that Jevons holds for software and will push employment up rather than down as AI penetrates the field. The framing is that agentic coding lowers the cost of building, and cheaper building unlocks a backlog of software the world always wanted but could not previously afford. Some early market signals point the same way, with demand for engineers holding or rising rather than collapsing into mass layoffs.

• The Counterintuitive Truth: AI Coding Agents Are Creating More Engineering Jobs, Not Fewer (Hatchworks)
• Jevons' Paradox and the Fate of Software Developers (Jim Rutt)

The caveats

The paradox is not a free lunch, and it does not promise universal upside.

• It requires efficiency gains where humans and AI are complements, not perfect substitution where AI alone does the whole job. If a role is fully substitutable, Jevons does not save it.

• Even under net expansion, some tasks get commoditized, some entry-level work becomes scarcer, some firms hold headcount flat while demanding more output, and there is real displacement in the transition even if the long run grows. That seam, where the entry-level rung thins, is exactly where the apprenticeship problem lives.

• AI, Jobs, and the Jevons Paradox: Why the Void Rarely Stays Empty (Hoyem) is a balanced take on displacement versus expansion.

The quality counter-signal

More software, produced faster, is not automatically more value. Industry research (see the DORA 2024 report) found that AI adoption raised individual productivity while being associated with a decline in software-delivery stability and throughput. That tempers a naive "expansion is pure good" read: the Jevons upside is real, but it lands only if the extra software is safe to run.