Who Prices Cognitive Labor in the Age of Agents? Compute-Anchored Wages
Pith reviewed 2026-05-11 00:52 UTC · model grok-4.3
The pith
Human wages for AI-substitutable cognitive tasks are capped by the compute rental cost of equivalent agent output.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Agents are a production technology converting compute capital K_c into effective units of cognitive labor L_A. On tasks where human and agent-produced cognitive labor substitute, the competitive human wage is therefore bounded above by lambda times k times r_c, with lambda the relative human-to-agent productivity, k the compute intensity per agent unit, and r_c the rental rate of compute. The bound is obtained by migrating the elastic-supply margin from the labor market to the compute market and is generalized via CES aggregation that distinguishes substitutes from complements, with resulting implications for factor shares.
What carries the argument
The Compute-Anchored Wage (CAW) bound, obtained by treating agents as compute-using technology in the factor-pricing framework and migrating the supply-elasticity margin to the compute capital market.
If this is right
- Human wages on substitutable tasks cannot exceed the compute cost of equivalent agent output.
- The elastic supply margin that determines cognitive labor prices migrates from labor markets to compute capital markets.
- CES aggregation separates the bound's applicability to substitute tasks from its absence on complementary tasks.
- Changes in compute rental rates or intensity directly transmit to human wage ceilings.
- Factor income shares adjust according to the relative productivity and intensity parameters.
Where Pith is reading between the lines
- If compute efficiency improves faster than agent productivity gains, the wage bound tightens even without changes in labor supply.
- Regulators interested in cognitive wage floors may need to target compute market structure rather than labor market rules.
- In the limit of perfect substitution and abundant compute, routine cognitive wages approach the marginal cost of additional compute.
- The framework suggests empirical tests that compare observed wages against measured compute costs on well-defined task categories.
Load-bearing premise
That human and agent outputs are close substitutes on the relevant tasks and that the compute capital market is competitive so the rental rate can be taken as given.
What would settle it
Direct evidence that wages for substitutable cognitive tasks remain well above lambda k r_c despite competitive compute rental rates would contradict the upper bound.
read the original abstract
A natural intuition about the economics of AI agents is that, because agents can be replicated at very low marginal cost, agent labor may be supplied highly elastically, placing downward pressure on cognitive-labor wages when it closely substitutes for human labor. We argue this framing is wrong in mechanism but partially correct in conclusion, and that the correction matters for both theory and policy. \textbf{Agents are not labor; they are a production technology that converts compute capital $K_c$ into effective units of cognitive labor $L_A$.} Once this is recognized, the elastic-supply margin that anchors the equilibrium wage migrates from the labor market to the compute capital market. Building on the classic factor-pricing framework \citep{mankiw2020}, we derive a \emph{Compute-Anchored Wage} (CAW) bound stating that, on tasks where human and agent-produced cognitive labor are substitutes, the competitive human wage is bounded above by $\lambda \cdot k \cdot r_c$, where $r_c$ is the rental rate of compute capital, $k$ is the compute intensity of one effective agent-produced cognitive labor unit, and $\lambda$ is the relative human-to-agent productivity. We generalize the result through constant elasticity of substitution (CES) aggregation, separate substitutable from complementary tasks, and discuss factor-share consequences. The conclusion is concise: \emph{the price-setter for cognitive labor is no longer the labor market.}
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper models AI agents as a production technology that converts compute capital K_c into effective cognitive labor units L_A rather than as direct labor substitutes. Building on standard factor-pricing models, it derives a Compute-Anchored Wage (CAW) bound: on tasks where human and agent cognitive labor are substitutes, the competitive human wage satisfies w_h ≤ λ · k · r_c, with λ the relative human-to-agent productivity, k the compute intensity per effective agent labor unit, and r_c the rental rate of compute. The central claim is that the elastic-supply margin anchoring wages therefore migrates from the labor market to the compute-capital market. The result is extended via CES aggregation to distinguish substitutable from complementary tasks and to discuss implications for factor shares.
Significance. If the derivation is robust to the market-structure assumptions, the CAW bound supplies a clean, parameter-light theoretical anchor for cognitive wages in AI-augmented economies. It shifts analytical focus from labor-supply elasticities to compute-market conditions and offers a transparent way to separate task types, which could inform both positive predictions about wage pressure and normative discussions of compute policy. The grounding in Mankiw-style production functions is a strength.
major comments (2)
- [CAW bound derivation and CES aggregation] CAW bound derivation (main text, following the production-function setup): The bound and the claim that 'the elastic-supply margin migrates to the compute capital market' treat r_c as parametric. The CES aggregation step does not specify whether compute supply is perfectly elastic or whether agent-driven demand is large enough to render r_c endogenous in general equilibrium. When cognitive labor accounts for a non-negligible share of total compute use, the human wage becomes jointly determined by labor and compute supply elasticities rather than being anchored solely by a fixed r_c; this directly affects the load-bearing claim that the price-setter is no longer the labor market.
- [CES aggregation and factor shares] Factor-share discussion (CES extension): The paper states that the result generalizes to separate substitutable and complementary tasks and discusses factor-share consequences, yet provides no explicit comparative-static expressions or illustrative parameter values showing how the human labor share changes with λ, k, or the elasticity of substitution. Without these, the quantitative implications for the central claim remain difficult to evaluate.
minor comments (1)
- [Abstract and introduction] The abstract and introduction could more explicitly list the maintained assumptions (competitive compute market, perfect substitutability on the relevant tasks) that are required for the bound to hold as stated.
Simulated Author's Rebuttal
We thank the referee for the constructive comments, which help clarify the scope and robustness of the Compute-Anchored Wage (CAW) framework. We respond to each major comment below and indicate the revisions we will make.
read point-by-point responses
-
Referee: [CAW bound derivation and CES aggregation] CAW bound derivation (main text, following the production-function setup): The bound and the claim that 'the elastic-supply margin migrates to the compute capital market' treat r_c as parametric. The CES aggregation step does not specify whether compute supply is perfectly elastic or whether agent-driven demand is large enough to render r_c endogenous in general equilibrium. When cognitive labor accounts for a non-negligible share of total compute use, the human wage becomes jointly determined by labor and compute supply elasticities rather than being anchored solely by a fixed r_c; this directly affects the load-bearing claim that the price-setter is no longer the labor market.
Authors: The CAW bound is obtained from firm cost minimization under competitive markets: the marginal cost of an effective unit of agent cognitive labor is k · r_c, so any human wage above λ · k · r_c would be dominated by agent use on substitutable tasks. This inequality is a partial-equilibrium no-arbitrage condition that continues to hold when r_c is endogenous; equilibrium r_c simply adjusts to clear the compute market, but the upper bound on w_h remains anchored by compute-market conditions rather than by human labor supply. The migration claim therefore refers to the relevant margin of adjustment (compute supply elasticity and any other determinants of r_c) rather than to a literally fixed r_c. We agree that the manuscript would be strengthened by an explicit discussion of the general-equilibrium case. In revision we will add a short subsection after the CES aggregation that derives the joint determination of w_h and r_c when agent demand is non-negligible and shows that the CAW bound is preserved while the quantitative wage pressure depends on compute supply elasticity. revision: yes
-
Referee: [CES aggregation and factor shares] Factor-share discussion (CES extension): The paper states that the result generalizes to separate substitutable and complementary tasks and discusses factor-share consequences, yet provides no explicit comparative-static expressions or illustrative parameter values showing how the human labor share changes with λ, k, or the elasticity of substitution. Without these, the quantitative implications for the central claim remain difficult to evaluate.
Authors: We accept the point. The current CES section separates task types and notes qualitatively that the human labor share falls with higher λ or lower k on substitutable tasks, but it stops short of closed-form comparative statics or numerical illustrations. In the revised manuscript we will (i) derive the explicit expression for the equilibrium human labor share under CES aggregation as a function of λ, k, the substitution elasticity σ, and the task weight, and (ii) include a brief table of illustrative values (e.g., σ = 1.5 and σ = 3 for substitutes; λ ranging from 0.5 to 2; k normalized to 1) that shows the resulting labor-share changes. These additions will make the quantitative implications transparent without altering the paper’s core theoretical result. revision: yes
Circularity Check
No significant circularity; derivation from standard factor pricing primitives
full rationale
The paper's central CAW bound is obtained by reinterpreting agents as a compute-capital-using production technology and applying the standard competitive factor-pricing condition from Mankiw (2020). The inequality w_h ≤ λ k r_c follows directly from cost minimization once the marginal cost of an effective agent unit is defined as k r_c and λ is introduced as the relative productivity scalar; neither λ nor k is calibrated to observed wage series, and r_c is treated as parametric under the maintained competitive-market assumption. No self-citations, fitted inputs renamed as predictions, or ansatz smuggling appear in the derivation chain. The CES generalization and task-separation steps are likewise direct applications of standard aggregation without circular reduction to the target result.
Axiom & Free-Parameter Ledger
free parameters (2)
- λ
- k
axioms (2)
- domain assumption Agents function as a production technology that converts compute capital K_c into effective cognitive labor L_A rather than as direct labor supply.
- standard math Human and agent cognitive labor are aggregated via constant elasticity of substitution on substitutable tasks.
invented entities (1)
-
Compute-Anchored Wage (CAW) bound
no independent evidence
Reference graph
Works this paper leans on
-
[1]
Daron Acemoglu and Pascual Restrepo. The race between man and machine: Implications of technology for growth, factor shares, and employment.American Economic Review, 108(6):1488–1542, 2018
work page 2018
-
[2]
Automation and new tasks: How technology displaces and reinstates labor
Daron Acemoglu and Pascual Restrepo. Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives, 33(2):3–30, 2019
work page 2019
-
[3]
Robots and jobs: Evidence from US labor markets.Journal of Political Economy, 128(6):2188–2244, 2020
Daron Acemoglu and Pascual Restrepo. Robots and jobs: Evidence from US labor markets.Journal of Political Economy, 128(6):2188–2244, 2020
work page 2020
-
[4]
Tasks, automation, and the rise in U.S
Daron Acemoglu and Pascual Restrepo. Tasks, automation, and the rise in U.S. wage inequality.Econometrica, 90(5):1973–2016, 2022
work page 1973
-
[5]
Philippe Aghion, Benjamin F. Jones, and Charles I. Jones. Artificial intelligence and economic growth. In Ajay Agrawal, Joshua Gans, and Avi Goldfarb, editors,The Economics of Artificial Intelligence: An Agenda, pages 237–282. University of Chicago Press, 2019
work page 2019
-
[6]
Katz, Christina Patterson, and John Van Reenen
David Autor, David Dorn, Lawrence F. Katz, Christina Patterson, and John Van Reenen. The fall of the labor share and the rise of superstar firms.Quarterly Journal of Economics, 135(2):645–709, 2020
work page 2020
-
[7]
David H. Autor. Why are there still so many jobs? the history and future of workplace automation.Journal of Economic Perspectives, 29(3):3–30, 2015
work page 2015
-
[8]
David H. Autor and David Dorn. The growth of low-skill service jobs and the polarization of the US labor market. American Economic Review, 103(5):1553–1597, 2013
work page 2013
-
[9]
Autor, Frank Levy, and Richard J
David H. Autor, Frank Levy, and Richard J. Murnane. The skill content of recent technological change: An empirical exploration. Quarterly Journal of Economics, 118(4):1279–1333, 2003
work page 2003
-
[10]
Timothy F. Bresnahan and Manuel Trajtenberg. General purpose technologies: “Engines of growth”?Journal of Econometrics, 65(1):83–108, 1995
work page 1995
-
[11]
Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond. Generative AI at work.NBER WorkingPaper No. 31161, 2023
work page 2023
-
[12]
The rising costs of training frontier ai models,
Ben Cottier, Robi Rahman, Loredana Fattorini, Nestor Maslej, and David Owen. The rising costs of training frontier AI models.arXiv preprint arXiv:2405.21015, 2024
-
[13]
Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock. GPTs are GPTs: An early look at the labor market impact potential of large language models.arXiv preprint arXiv:2303.10130, 2023
-
[14]
Felten, Manav Raj, and Robert Seamans
Edward W. Felten, Manav Raj, and Robert Seamans. How will language modelers like ChatGPT affect occupations and industries? arXiv preprint arXiv:2303.01157, 2023
-
[15]
Avi Goldfarb, Bledi Taska, and Florenta Teodoridis. Could machine learning be a general purpose technology? a comparison of emerging technologies using data from online job postings.Research Policy, 52(1):104653, 2023
work page 2023
-
[16]
Katz.The Race Between Education and Technology
Claudia Goldin and Lawrence F. Katz.The Race Between Education and Technology. Harvard University Press, 2008
work page 2008
-
[17]
The global decline of the labor share.Quarterly Journal of Economics, 129(1):61–103, 2014
Loukas Karabarbounis and Brent Neiman. The global decline of the labor share.Quarterly Journal of Economics, 129(1):61–103, 2014
work page 2014
-
[18]
Lawrence F. Katz and Kevin M. Murphy. Changes in relative wages, 1963–1987: Supply and demand factors. Quarterly Journal of Economics, 107(1):35–78, 1992
work page 1963
-
[19]
Language models and cognitive automation for economic research
Anton Korinek. Language models and cognitive automation for economic research. NBER Working Paper No. 30957, 2023
work page 2023
-
[20]
Anton Korinek and Joseph E. Stiglitz. Artificial intelligence and its implications for income distribution and unemployment. In Ajay Agrawal, Joshua Gans, and Avi Goldfarb, editors,The Economics of ArtificialIntelligence: An Agenda, pages 349–390. University of Chicago Press, 2019
work page 2019
-
[21]
Ohanian, José-Víctor Ríos-Rull, and Giovanni L
Per Krusell, Lee E. Ohanian, José-Víctor Ríos-Rull, and Giovanni L. Violante. Capital-skill complementarity and inequality: A macroeconomic analysis.Econometrica, 68(5):1029–1053, 2000
work page 2000
-
[22]
Gregory Mankiw.Principles of Economics
N. Gregory Mankiw.Principles of Economics. Cengage Learning, 9 edition, 2020
work page 2020
-
[23]
Experimental evidence on the productivity effects of generative artificial intelligence
Shakked Noy and Whitney Zhang. Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654):187–192, 2023. 10
work page 2023
-
[24]
Harvard University Press, 2014
Thomas Piketty.Capital in the Twenty-FirstCentury. Harvard University Press, 2014
work page 2014
-
[25]
Jaime Sevilla, Lennart Heim, Anson Ho, Tamay Besiroglu, Marius Hobbhahn, and Pablo Villalobos. Compute trends across three eras of machine learning.2022 International Joint Conference on Neural Networks (IJCNN), 2022
work page 2022
-
[26]
Daniel Susskind.A WorldWithout Work: Technology,Automation, and How WeShould Respond. Metropolitan Books, 2020
work page 2020
-
[27]
Economic growth under transformative AI.Annual Review of Economics, 15:567–593, 2023
Philip Trammell and Anton Korinek. Economic growth under transformative AI.Annual Review of Economics, 15:567–593, 2023. 11 Appendix A Notation Summary For convenience we collect the symbols used throughout the paper. •Y: aggregate output of the representative competitive firm. •K,L: generic physical capital and labor in the Mankiw [22] setup. •K o: non-c...
work page 2023
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.