Paradoxe de Jevons

William Stanley Jevons observed in 1865 that greater efficiency raises consumption rather than cutting it. Applied to software, the claim is sharper: if the price elasticity of demand for code exceeds 1, cheaper production raises the volume demanded more than proportionally. The people who invoke it disagree about what follows.

The economic framing comes from Trésor-Éco n° 391 (DG Trésor, June 2026): exposure to automation does not determine employment, price elasticity does. A job that is substitutable but faces elastic demand can see employment grow. SFEIR, commenting on that note on 23 July 2026, pushes the point further: developers and graphic designers sit above an elasticity of 1, so the argument is structurally pro-employment for developers. Simon Wardley adds a Red Queen effect on 27 March 2026: competing firms reinvest every productivity gain, so a large company moves from 30 million lines of code to over a billion just to hold position. Philippe Ensarguet states the limit case on 11 March 2026, when execution cost falls to near zero and demand becomes infinite, with budget lines migrating from salaries toward compute.

Dave Farley reads the same curve as a liability. In his 13 May 2026 note, more code means more complexity, more integration points, more behavior to evaluate, more maintenance, and less time left to understand the problem. He calls AI-assisted development without continuous delivery a complexity bomb with a delayed fuse.

So Jevons arrives in two incompatible shapes: an argument that developer demand expands, and an argument that what expands is the maintenance surface.