🔘 About this critique
A model may outperform an individual worker and still struggle to replace a well-organized firm. Organizing Intelligence makes that distinction the starting point of an economic theory of AI. Its central question is how individual capabilities combine: stronger task performance and better coordination can change organizations in different ways, with different consequences for employment.
This assessment examines arXiv version 1, including its proofs and numerical-method appendices. Selected illustrative calculations were checked independently. It is an editorial assessment of a theoretical preprint, not peer review, empirical validation or a complete replication of the numerical analyses.
Critique
| Examined version | arXiv:2611.05417v1, submitted on 31 October 2026. Theoretical preprint; no peer-reviewed journal version is identified in the examined record. |
| Materials examined | arXiv record and complete 108-page manuscript, including Appendices A–E. |
| Recommendation | Read Propositions 3 and 5 with their assumptions, Figure 23 with Appendix D.4, and the numerical qualifications in Appendices D.7 and E.6. |
Unit 1 — What would count as a coordination breakthrough?
Article data. The model separates competence, which affects productivity levels, from coordination, which governs the returns to organizational scale. Assumption 1 imposes an asymptotic power law for output as a fixed workforce composition expands. Proposition 3 distinguishes a machine-only exponent below one from an exponent above one. In the illustrative CES technology, coordination equal to one gives a finite limiting output per agent; only values strictly above one make that quantity unbounded.
Reading. This is a meaningful distinction between solving tasks and combining contributions. However, the exponent is a property of an assumed production technology, not an observed score on a general intelligence test. The literature discussed in Section 7.2 provides motivating evidence; it does not estimate the complete economic model. The authors themselves propose additional measurements of collective output, computational expenditure and organizational composition in Section 8.
Critical conclusion. The paper identifies a mechanism that existing individual-task benchmarks can miss. Applying its threshold to a real system would require an economic measure of output, an explicit treatment of compute and communication costs, and evidence that the estimated scaling relationship persists beyond the tested team sizes. The preprint does not establish when that threshold will be crossed.
| Claim type | Conditional scaling result and empirical interpretation |
| Exposure | High. Treating a theoretical exponent as a measured capability threshold would turn a conditional mechanism into an unsupported forecast. |
Unit 2 — Concentration under an efficient replacement process
Article data. Proposition 3 proves concentration in the limit as demand scale tends to infinity, with human resources fixed and identical AI copies available at a constant rental cost. Equilibrium also requires replacement stability: an alternative organization can challenge the entire industry using transfers that preserve the payoffs of consumers and human resource owners. Appendix B allows divisible activity weights, so the concentration index measures complete-firm capacities rather than necessarily counting an integer number of businesses.
Reading. The result explains why sufficiently strong scale economies favour concentrating production. Its institutional interpretation is narrower than a prediction of monopoly power. Firms earn zero profits at supporting prices, and the replacement process selects total-surplus efficiency. These premises do not establish how an incumbent with entry barriers would price its product or divide income. The authors acknowledge that regulation, rival providers, capacity constraints and supply contracts could obstruct extreme concentration.
Critical conclusion. The theorem is useful for identifying a pressure towards concentration under specified technological and institutional conditions. It does not demonstrate inevitable monopoly, monopoly rents or a corresponding distribution of household income. Reported wages must also be distinguished from final payoffs after the model’s transfers.
| Claim type | Market structure, equilibrium institutions and distribution |
| Exposure | High. Efficient concentration in this model cannot by itself establish the market power or welfare effects of an actual dominant supplier. |
Unit 3 — Why employment can decline gradually or abruptly
Article data. Proposition 5 studies one human type and the best output per agent attainable at each human share. With a strictly concave relationship and the stated feasibility conditions, employment changes continuously as AI capabilities improve. With strict convexity, feasible pure-production scales and active machine firms, production separates into human-only and machine-only firms; employment can then switch discontinuously when their unit costs cross. Appendix D.4 establishes the two examples in Figure 23 analytically.
Reading. The employment discontinuity does not require crossing the large-market coordination threshold. In Figure 23’s second example, AI coordination remains fixed at 0.9999 while task competence increases. Recalculating the stated cost equality gives a competence threshold of approximately 1.0796369. At the crossing, multiple employment levels are compatible with equilibrium. These are dimensionless model quantities, not measured probabilities, dates or benchmark targets.
Critical conclusion. The distinction between replacing workers within firms and replacing the firms that employ them is a substantial contribution. It specifies conditions under which smooth technical progress can produce a discontinuous employment response. The model does not estimate the speed of adjustment in an economy with retraining, hiring frictions, changing human skills or changing demand.
| Claim type | Conditional employment discontinuity and comparative statics |
| Exposure | High. A jump between equilibria should not be interpreted as an estimated timetable or probability of mass unemployment. |
Unit 4 — The evidential status of the numerical illustrations
Article data. The appendices distinguish analytic results from numerical checks. For Figure 22, Appendix D.5 reports entry-profit bounds at 99 sampled AI profiles, with all compositions and scales covered at those points. By contrast, the finite-market curves in Figures 17 and 26 optimize over pure firms, supplemented by entry searches at selected states. Appendix D.7 explicitly states that these checks do not establish equilibrium along the full paths. It also describes Figure 25’s wage comparison as descriptive rather than an application of the interior-match derivative formula throughout the path.
Reading. These qualifications materially improve the transparency of the paper. They also determine how its figures should be read. A smooth plotted line can connect calculations with different degrees of verification. The analytic large-market concentration theorem and limiting wage result have separate proofs; incomplete certification of a finite-market illustration does not refute those proofs. Conversely, a theorem with specified conditions does not automatically certify every displayed numerical path.
Critical conclusion. Use the analytic examples to understand mechanisms and retain the stated limits of the numerical illustrations. The manuscript supports a structured programme of empirical and computational investigation. The selected calculations checked here do not constitute an independent reproduction of every equilibrium or proof.
| Claim type | Scope of numerical verification and reproducibility |
| Exposure | Moderate. The authors disclose the limitations, but treating every plotted path as fully established would overstate the computational evidence. |
Overall assessment
The paper’s strongest contribution is to place the organization between AI capabilities and economic outcomes. Complementary workers can form a productive combination that an individual-level comparison misses; improved AI coordination can alter which combinations remain viable. The analysis also explains why changes in wages may reflect the reorganization of employers as well as direct substitution by machines. Its connections to the conventional task model help make these additional mechanisms explicit.
Organizing Intelligence is recommended as a theoretical framework and a guide to measurement. Its conclusions are conditional on production, supply, demand and equilibrium assumptions. The next evidential step is to measure collective performance and organizational change in settings where costs, human adaptation and institutional constraints can be observed. Read the claims about concentration and employment together with their propositions and appendices, rather than treating the numerical scenarios as economic forecasts.
🔘 Paper page: https://arxiv.org/abs/2611.05417v1
Abstract
When will advances in AI lead to economic transformation? We develop a model of how intelligence, the ability to achieve goals, matters for organizations, the key unit of economic production. Agents (human or AI) with multidimensional capabilities sort endogenously into competing firms. A firm’s productivity depends on its agents’ capabilities; we distinguish competence, which shifts productivity levels, from coordination, which governs how productivity scales with size. Two differences between AI and humans drive transformation: AI expertise is reproducible while humans’ are in fixed supply, and AIs may not face the same human limits on coordination. We analyze how equilibrium market structure, productivity, employment, and wages evolve as AI capabilities improve. When AI is more capable than the average human, organizations of complementary humans can nonetheless outperform AI. Increases in AI competence can change the composition of workers within firms and decrease firm size. As AI coordination improves, however, this can break the Coasean ceiling on firm size and the market structure undergoes a phase transition toward large, concentrated AI-heavy firms. Changes in human employment are gradual when AI displaces workers within mixed human-AI firms, or sudden when AI firms outcompete human firms. We decompose wage changes into a technology effect, which changes the value of humans within existing firms, and a reorganization effect as firms adjust size and composition. Reorganization can reverse the technology effect on wages.
Authors’ abstract reproduced from arXiv version 1 under the Creative Commons Attribution 4.0 International licence. Authors: Alex Imas, Andrew Koh and Shane Legg (2026).
Authors
Alex Imas, Andrew Koh, Shane Legg
Liked this post? Follow this blog to get more.