Artificial Intelligence in Team Dynamics: Who Gets Replaced and Why?

Mustafa Dogan; Pinar Yildirim; Xienan Cheng

arxiv: 2506.12337 · v3 · pith:HWJUDDTLnew · submitted 2025-06-14 · 💰 econ.TH

Artificial Intelligence in Team Dynamics: Who Gets Replaced and Why?

Xienan Cheng , Mustafa Dogan , Pinar Yildirim This is my paper

Pith reviewed 2026-05-25 07:49 UTC · model grok-4.3

classification 💰 econ.TH

keywords AI adoptionteam productionpeer monitoringmoral hazardsequential workflowwage inequalityoptimal replacement

0 comments

The pith

In sequential team workflows, optimal AI replaces workers at the start and end but spares the middle one to preserve peer monitoring signals.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper builds a model of team production in which a principal disciplines workers through peer monitoring, where each observes a signal about the predecessor's effort. It shows that limited AI should be deployed stochastically at the workflow's beginning and end positions, leaving the middle worker human because that position transmits the monitoring information. The principal may also choose to leave some AI capacity unused rather than replace as many workers as possible. Under this strategy, average wages rise and wage differences within the team shrink.

Core claim

The optimal AI strategy stochastically replaces workers at the beginning and at the end of the workflow, but does not replace the middle worker, since this worker is crucial for sustaining the flow of information obtained by peer monitoring. The principal may optimally underutilize available AI capacity. The optimal AI adoption increases average wages and reduces intra-team wage inequality.

What carries the argument

Sequential team production model with peer monitoring signals whose value depends on retaining a human worker in the middle position.

If this is right

Workers at the workflow start and end face the highest replacement risk under optimal AI use.
The middle worker position is protected because it sustains the flow of monitoring information.
The principal may leave some available AI capacity idle to maintain discipline.
Average wages across the team increase.
Wage differences within the team decrease.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Firms organized in linear sequences may retain human coordinators even when parallel tasks are more readily automated.
The pattern implies different AI risks for information-hub roles versus peripheral ones in monitored production.
Real-world tests could compare AI adoption rates in sequential versus independent task teams to check the monitoring channel.
Wage compression suggests monitored teams may see inequality effects distinct from those in unmonitored settings.

Load-bearing premise

Peer monitoring generates a usable signal whose value depends on the middle worker remaining human, while AI agents face no moral hazard.

What would settle it

An empirical setting or experiment in which replacing the middle worker leaves monitoring signals and effort levels unchanged, or in which AI agents exhibit moral hazard comparable to humans, would falsify the predicted replacement pattern.

read the original abstract

This study investigates the effects of artificial intelligence (AI) adoption in organizations. We ask: First, how should a principal optimally deploy limited AI resources to replace workers in a team? Second, in a sequential workflow, which workers face the highest risk of AI replacement? Third, would the principal always prefer to fully utilize all available AI resources, or are there any benefits to keeping some slack AI capacity? Fourth, what are the effects of optimal AI deployment on the wage level and intra-team wage inequality? We develop a sequential team production model in which a principal can use peer monitoring--where each worker observes a signal of their predecessor's effort--to discipline team members. The principal may replace some workers with AI agents, whose actions are not subject to moral hazard. Our analysis yields four key results. First, the optimal AI strategy stochastically replaces workers rather than fixating on a single position. Second, the principal replaces workers at the beginning and at the end of the workflow, but does not replace the middle worker, since this worker is crucial for sustaining the flow of information obtained by peer monitoring. Third, the principal may optimally underutilize available AI capacity. Fourth, the optimal AI adoption increases average wages and reduces intra-team wage inequality.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The model claims optimal AI stochastically replaces workflow ends but spares the middle worker to preserve peer monitoring signals, plus possible underutilization of AI capacity.

read the letter

The main thing to know is that this paper builds a sequential team model where the principal uses peer monitoring to handle moral hazard and then chooses which positions to replace with AI agents that have no moral hazard. It concludes that replacement hits the first and last spots stochastically but leaves the middle human, that full AI use is not always optimal, and that this raises average wages while cutting inequality inside the team. The specific combination of endogenous position choice with the monitoring chain and the underutilization result looks like the freshest part relative to standard team-theory setups. The framing around the four questions is clean and directly connects replacement risk to information flow. The soft spot is that the abstract alone gives the four results without the signal technology, equilibrium definition, or incentive constraints, so the stress-test concern lands: the middle-worker protection appears to rest on the assumption that only a human in that spot keeps the monitoring signal usable downstream. If the paper does not show this holds under alternative monitoring structures or if AI can relay signals, the replacement pattern and underutilization claim become fragile. The wage and inequality effects are downstream of the same unshown derivations. This is for readers in organizational economics or contract theory who work on internal firm structure and technology. A serious referee could check whether the results are robust or tied to particular functional forms, so the paper deserves that step even with the current gaps in the abstract.

Referee Report

2 major / 1 minor

Summary. The paper develops a sequential team production model in which a principal can replace some workers with AI agents that face no moral hazard. Workers use peer monitoring, with each observing a signal of their predecessor's effort. The analysis yields four results: optimal AI deployment is stochastic rather than position-specific; the principal replaces workers at the beginning and end of the workflow but not the middle worker (who sustains the information flow); the principal may optimally underutilize available AI capacity; and optimal AI adoption raises average wages while reducing intra-team wage inequality.

Significance. If the central claims hold under the model's assumptions, the paper contributes to organizational economics by showing how information externalities from sequential peer monitoring create position-dependent replacement risks and can make partial AI adoption optimal. The wage and inequality predictions offer testable implications for how AI affects team compensation structures.

major comments (2)

[Abstract] Abstract (and model description): The claim that the middle worker is never replaced follows from the assumption that peer-monitoring signals are generated and transmitted only when the middle position remains human. Without the explicit functional form of the signal, the incentive-compatibility constraints, or the equilibrium conditions shown, it is impossible to verify whether this replacement pattern is robust or an artifact of the specific monitoring technology.
[Abstract] Abstract (results on underutilization): The result that the principal may optimally underutilize AI capacity rests on the moral-hazard distinction between humans and AI. If the monitoring technology allows AI to produce or relay usable signals (or if monitoring is not strictly predecessor-only), both the stochastic end-only replacement pattern and the underutilization finding collapse.

minor comments (1)

[Abstract] The abstract states the four results but provides no indication of the number of periods, the production function, or the distribution of the monitoring signal, which would aid readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for these constructive comments, which highlight important scope conditions of the model. We respond to each major comment below and propose targeted revisions to improve clarity.

read point-by-point responses

Referee: [Abstract] Abstract (and model description): The claim that the middle worker is never replaced follows from the assumption that peer-monitoring signals are generated and transmitted only when the middle position remains human. Without the explicit functional form of the signal, the incentive-compatibility constraints, or the equilibrium conditions shown, it is impossible to verify whether this replacement pattern is robust or an artifact of the specific monitoring technology.

Authors: Section 2 of the manuscript defines the monitoring technology explicitly as predecessor-only: worker i receives signal s_i = e_{i-1} + ε_i (with ε_i ~ N(0,σ²)) only when position i-1 is human. AI agents have no effort choice and generate no signal. The incentive-compatibility constraints appear as inequalities (3)–(5) in Section 3, and the principal’s program is solved in Section 4, where replacing the middle position severs the chain and violates IC for the terminal worker. The result is therefore tied to this technology, which we view as a natural representation of sequential peer monitoring. We will add the key signal equation and a footnote to the abstract, plus a short robustness paragraph in the introduction, in the revised version. revision: yes
Referee: [Abstract] Abstract (results on underutilization): The result that the principal may optimally underutilize AI capacity rests on the moral-hazard distinction between humans and AI. If the monitoring technology allows AI to produce or relay usable signals (or if monitoring is not strictly predecessor-only), both the stochastic end-only replacement pattern and the underutilization finding collapse.

Authors: Under the paper’s maintained assumption of strictly predecessor-only monitoring by humans, AI cannot substitute for the middle worker’s role in transmitting the signal that disciplines the team; hence the principal may leave AI capacity idle. The manuscript does not claim the result holds for every conceivable monitoring technology. We will insert an explicit caveat in the introduction and a new subsection 5.3 discussing how the findings would change if AI could generate or relay signals, while preserving the core results under the stated assumptions. revision: partial

Circularity Check

0 steps flagged

No circularity: theoretical model results derived from stated assumptions without reduction to inputs or self-citations

full rationale

The abstract describes a sequential team production model with peer monitoring and moral hazard distinctions, from which four results are claimed to follow via analysis. No equations, fitted parameters, or self-citations are referenced in the provided text. The replacement pattern, underutilization, and wage effects are presented as outputs of the model rather than inputs or renamings. The derivation chain is therefore self-contained against external benchmarks, with the monitoring technology and incentive constraints serving as independent primitives. No load-bearing step reduces by construction to a fit or prior author result.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the modeling choices that peer monitoring transmits information only when the middle worker is human and that AI agents are immune to moral hazard; these are domain assumptions introduced to generate the replacement pattern.

axioms (2)

domain assumption Peer monitoring generates a signal whose disciplinary value depends on the middle worker remaining human.
Invoked to justify sparing the middle position; stated in the model description in the abstract.
domain assumption AI agents face no moral hazard while human workers do.
Core modeling distinction that enables replacement without incentive costs; stated in the abstract.

pith-pipeline@v0.9.0 · 5753 in / 1366 out tokens · 34099 ms · 2026-05-25T07:49:29.611709+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We develop a sequential team production model in which a principal can use peer monitoring—where each worker observes a signal of their predecessor's effort—to discipline team members. The principal may replace some workers with AI agents, whose actions are not subject to moral hazard.
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

the optimal AI strategy stochastically replaces workers at the beginning and at the end of the workflow, but does not replace the middle worker, since this worker is crucial for sustaining the flow of information obtained by peer monitoring.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

12 extracted references · 12 canonical work pages

[1]

Trigger strategy profile σ∗ constitutes an equilibrium under wx

work page
[2]

No alternative compensation scheme with a lower expected cost can support σ∗ as an equilibrium

work page
[3]

Supporting any other strategy profile, apart from the trigger strategy profile, that results in joint effort, as an equilibrium necessarily incurs a higher expected compensation cost for the principal. Step 1. We will prove that, under the compensation scheme wx, the trigger strategy is a best response for each worker i ∈ {1, . . . , n} given that all oth...

work page
[4]

Then, a replacement strategy x = ( x1, 0, x3) with x1 > x 3 is optimal

Suppose not for contradiction. Then, a replacement strategy x = ( x1, 0, x3) with x1 > x 3 is optimal. This in turn requires x1 ∈ (0, 1). The expected compensation cost of the principal under the scheme ( x1, 0, x3) is: c x1 + x3 + p3 (1 − x1)2 (1 − x1)(p3 − p0) − x3(p1 − p0) + 1 (p3 − p1) − x3(p2 − p1) + 1 − x3 p3 − p2 . The expected compensation cost of...

work page
[5]

If p2 1 − p3p0 ≤ 0, the principal fully utilizes the AI capacity, i.e., ¯x∗ = 1

work page
[6]

If p2 1 − p3p0 > 0, the principal does not fully utilize the AI capacity, i.e., ¯x∗ < 1. Step 1. Assume that p2 1 − p3p0 ≤ 0. Suppose, for contradiction, that the principal does not fully utilize the AI capacity under the optimal replacement strategy, i.e., a strategy x with ¯x < 1 is optimal. Note that ζ1 = p0 + 3X k=2 xk 1 − x1 (p4−k − p0) . As we know ...

work page
[7]

First, x∗ 2 = 0 implies ζ x∗ 1 ≤ p1

From earlier results, we know that x∗ 3 ∈ (0, 1), and x∗ 2 = 0. First, x∗ 2 = 0 implies ζ x∗ 1 ≤ p1. Next, x∗ 3 ∈ (0, 1) implies ζ x∗ 2 ∈ (p1, p2). In consequence, we have ζ x∗ 3 > ζ x∗ 2 > ζ x∗ 1 , which implies that the wages are increasing in the position after the optimal automation: wx∗ 3 > w x∗ 2 > w x∗ 1

work page
[8]

As x∗ 3 > 0, we have ζ x∗ 1 > p 0 = ζ0

work page
[9]

This implies that the wage of the front-end worker increases after the automation: wx∗ 1 > w0 1

work page
[10]

x∗ 3 > 0 also implies that ζ x∗ 2 > p 1 = ζ0

work page
[11]

Therefore, the wage of the middle worker also increases after the automation: wx∗ 2 > w0 2

work page
[12]

Therefore, the wage of the end-most worker remains the same after the automation: wx∗ 3 = w0 3

Finally, note that, regardless of the replacement strategy, we have: ζ x∗ 3 = p2. Therefore, the wage of the end-most worker remains the same after the automation: wx∗ 3 = w0 3. Proof of Proposition 6 . From Proposition 5, we know that after AI replacement, the wages of the front-most and middle workers increase, while the wage of the end-most worker rema...

work page

[1] [1]

Trigger strategy profile σ∗ constitutes an equilibrium under wx

work page

[2] [2]

No alternative compensation scheme with a lower expected cost can support σ∗ as an equilibrium

work page

[3] [3]

Supporting any other strategy profile, apart from the trigger strategy profile, that results in joint effort, as an equilibrium necessarily incurs a higher expected compensation cost for the principal. Step 1. We will prove that, under the compensation scheme wx, the trigger strategy is a best response for each worker i ∈ {1, . . . , n} given that all oth...

work page

[4] [4]

Then, a replacement strategy x = ( x1, 0, x3) with x1 > x 3 is optimal

Suppose not for contradiction. Then, a replacement strategy x = ( x1, 0, x3) with x1 > x 3 is optimal. This in turn requires x1 ∈ (0, 1). The expected compensation cost of the principal under the scheme ( x1, 0, x3) is: c x1 + x3 + p3 (1 − x1)2 (1 − x1)(p3 − p0) − x3(p1 − p0) + 1 (p3 − p1) − x3(p2 − p1) + 1 − x3 p3 − p2 . The expected compensation cost of...

work page

[5] [5]

If p2 1 − p3p0 ≤ 0, the principal fully utilizes the AI capacity, i.e., ¯x∗ = 1

work page

[6] [6]

If p2 1 − p3p0 > 0, the principal does not fully utilize the AI capacity, i.e., ¯x∗ < 1. Step 1. Assume that p2 1 − p3p0 ≤ 0. Suppose, for contradiction, that the principal does not fully utilize the AI capacity under the optimal replacement strategy, i.e., a strategy x with ¯x < 1 is optimal. Note that ζ1 = p0 + 3X k=2 xk 1 − x1 (p4−k − p0) . As we know ...

work page

[7] [7]

First, x∗ 2 = 0 implies ζ x∗ 1 ≤ p1

From earlier results, we know that x∗ 3 ∈ (0, 1), and x∗ 2 = 0. First, x∗ 2 = 0 implies ζ x∗ 1 ≤ p1. Next, x∗ 3 ∈ (0, 1) implies ζ x∗ 2 ∈ (p1, p2). In consequence, we have ζ x∗ 3 > ζ x∗ 2 > ζ x∗ 1 , which implies that the wages are increasing in the position after the optimal automation: wx∗ 3 > w x∗ 2 > w x∗ 1

work page

[8] [8]

As x∗ 3 > 0, we have ζ x∗ 1 > p 0 = ζ0

work page

[9] [9]

This implies that the wage of the front-end worker increases after the automation: wx∗ 1 > w0 1

work page

[10] [10]

x∗ 3 > 0 also implies that ζ x∗ 2 > p 1 = ζ0

work page

[11] [11]

Therefore, the wage of the middle worker also increases after the automation: wx∗ 2 > w0 2

work page

[12] [12]

Therefore, the wage of the end-most worker remains the same after the automation: wx∗ 3 = w0 3

Finally, note that, regardless of the replacement strategy, we have: ζ x∗ 3 = p2. Therefore, the wage of the end-most worker remains the same after the automation: wx∗ 3 = w0 3. Proof of Proposition 6 . From Proposition 5, we know that after AI replacement, the wages of the front-most and middle workers increase, while the wage of the end-most worker rema...

work page