There’s a fundamental tension between the top-down tool-to-work model foundational in economics and the bottom-up work-to-tool model we see across other disciplines—a tension that Mokyr’s recent Nobel highlights.
The Economic Orthodoxy: Tool-to-Work
Tool-to-work assumes we invent things, then apply them to work, resulting in productivity. Think Mokyr’s epistemic base → propositional knowledge → prescriptive knowledge → profitable technique. Though Mokyr’s framework does include tacit craft knowledge in “prescriptive knowledge,” the flow still runs from knowledge first to application second.
This assumption is mathematically enforced throughout economics:
1. Solow residual: A ≡ Y / F(K,L)
Measured TFP is defined as the unexplained bit after capital and labour. By construction it is assumed to arrive from outside the production function—manna from heaven.
2. Production function: Y = A(t)K^α L^(1-α)
A(t) is exogenous and autonomous. The algebra literally prevents A from being produced inside the firm by problem-solving workers.
3. Romer: Ḋ = δL_A A^φ
Ideas are created by dedicated research labour L_A paid to do patentable R&D. This assumes innovation comes from dedicated research labor, but most innovation actually comes from productive labor solving immediate problems. The R&D lab model is the exception, not the rule.
4. Growth accounting: ΔTFP ≡ ΔY − αΔK − (1−α)ΔL
When a foreman’s workaround raises output without measurable capital or labour, the formula must credit an exogenous tech shock—even though the workaround was endogenous labour. The math forces a false inference: it doesn’t just remain agnostic about causation, it actively obscures it.
The Empirical Reality: Work-to-Tool
But when we look outside economics, the causal arrow points the other way: work → tool.
Pasquinelli, Matteo – The Eye of the Master (2023)
Labour-theoretic account: every AI model is “congealed labour” extracted from human feedback loops.
Hutchins, Edwin – Cognition in the Wild (1995)
How a cockpit (or a ship’s bridge) “remembers” speeds through division of labour; technology is the residue of reorganised work.
Nelson & Winter – An Evolutionary Theory of Economic Change (1982)
Firms carry “routines” that co-evolve with technology; growth is selection over tacit variations.
von Hippel, Eric – The Sources of Innovation (1988) / Democratizing Innovation (2005)
Users, not manufacturers, generate the key functional prototypes that are later codified.
Rosenberg, Nathan – Inside the Black Box (1982)
“Learning by using”: performance characteristics only become clear through actual deployment; the tool is refined by the work.
Orr, Julian – Talking about Machines (1996)
Photocopier technicians’ informal knowledge-sharing; the service manual lags what the work-community already knows.
Edgerton, David – The Shock of the Old (2006)
“Technology-in-use” perspective: innovation is maintenance, repair and incremental re-deployment, not invention.
Bahk & Gort – “Decomposing Learning by Doing” (JPE 1993)
Plant-level panel data: 50% of TFP growth comes from plant-specific learning, not R&D or patent spill-overs.
Evidence from General Purpose Technologies
When you survey innovations with significant economic impact—think General Purpose Technologies, though that’s more a post-fact rationalisation than predictive analytic—the vast majority follow work-to-tool:
Steam power: Newcomen’s engine emerged from solving mining pumping problems. Decades of incremental improvements by millwrights, engineers, and operators—people doing the work—transformed it into a general-purpose technology.
Flight: The Wright brothers were solving a control problem in existing glider work. They weren’t applying aerodynamic theory; they were trying to stay aloft, and the control surfaces crystallized from that struggle.
Linux and open source: Millions of incremental commits from users/developers solving their own problems eventually congeal into “technology.” No one designed Linux top-down; it emerged from the work.
The Implication
If technology is congealed work rather than applied science, then productivity policy focused on R&D subsidies is systematically missing the action.
The real question becomes: what work conditions allow productive crystallization?
Not: how do we generate more patentable ideas?
But: how do we organize work so that the foreman’s workaround, the technician’s fix, the user’s modification—all that endogenous problem-solving—can accumulate and diffuse rather than dissipate?