Baumol’s cost disease

William Baumol’s famous example: it took four musicians about 40 minutes to play a Beethoven string quartet in 1826, and it takes four musicians about 40 minutes today. Zero productivity growth in two centuries — by design; that’s what the piece is.

Yet those musicians are paid vastly more than their 1826 counterparts. Why? Because they could otherwise take jobs in sectors where productivity did explode. To keep a cellist out of a high-productivity job, the orchestra has to pay something close to what that job would pay. Wages in stagnant sectors are dragged up by productivity gains in booming ones.

The engine is simple once you see it:

• Some sectors get rapid productivity growth (manufacturing, software). Output per worker soars, so wages there can rise without raising prices.
• Workers are mobile across sectors, so wages must rise everywhere to retain them — including in sectors with no productivity growth.
• In the stagnant sectors, rising wages with flat output-per-worker means rising unit costs, so their prices climb relative to everything else.

This is why healthcare, education, live performance, and in-person care take an ever-larger share of spending over time. It isn’t mismanagement — it’s the mechanical consequence of uneven productivity growth. Hence “disease,” though Baumol stressed it’s a sign of wealth, not decay: we can afford the expensive services precisely because the other sectors got so productive.

This is the Pere reason it’s a primitive. AI is an enormous, uneven productivity shock — it tears through verifiable digital domains (code, content, analysis, structured extraction) while barely touching precision-critical, low-verifiability, physical, and high-judgment work. That unevenness is exactly the input Baumol’s model takes.

Run the mechanism forward: as AI makes the automatable domains radically more productive, the relative cost — and relative value — of the domains it can’t cheaply do rises. The work that ends up scarce and well-paid is precisely the work on the wrong side of the jagged edge: a nurse’s hands, a plumber’s site visit, a negotiation, a courtroom argument, care, taste, accountability.

So Baumol is the economic shadow of jagged intelligence. Jevons tells you the cheapened work expands; Baumol tells you human labour concentrates in — and is repriced upward for — what AI is bad at.

When forecasting AI’s effect on a sector or a wage, ask the Baumol questions:

1. How much can AI raise productivity here? Verifiable and digital → a lot. Precision-critical, physical, or judgment-heavy → little.
2. What happens to relative price? Low-automatability sectors get relatively more expensive over time, even as the high-automatability ones deflate.
3. Where does that push wages and talent? Toward the scarce human capabilities AI can’t replicate — the “cost disease” jobs become the resilient, high-value ones.

Read alongside Jevons Paradox and the lump-of-labour fallacy. The three together: work doesn’t run out (lump-of-labour), cheapened work expands (Jevons), and the human share migrates to — and is repriced up in — what stays hard (Baumol).