AI as Coordination-Compressing Capital: Task Reallocation, Organizational Redesign, and the Regime Fork

This paper proposes a model of "agent capital" where AI reduces coordination costs to endogenously reshape organizational structures, revealing a regime fork where the distributional outcome—ranging from broad-based gains to superstar concentration—depends on who controls the resulting organizational elasticity, even as overall inequality declines through employment expansion.

The Gist

Imagine a bustling newsroom. You have a Chief Editor at the top, a few Section Editors in the middle, and dozens of Reporters at the bottom.

In the old days, the Chief Editor could only manage three Section Editors. Each Section Editor could only manage three Reporters. Why? Because managing people is hard. It takes time to give instructions, check work, and fix mistakes. This "friction" limits how big a team can get before it becomes a mess.

Now, imagine a new super-tool arrives: AI. But not just AI that writes articles for you. This is AI that helps managers manage. It acts like a super-assistant that handles the boring coordination stuff, so a manager can talk to 20 people instead of just 3.

This paper, written by Alex Farach, asks a simple but huge question: What happens when AI makes it easier for managers to run bigger teams?

Here is the breakdown in everyday language, using some creative metaphors.

1. The Core Idea: The "Traffic Jam" Solution

Think of a company like a highway.

• The Problem: Usually, there's a traffic jam in the middle. The "managers" are the toll booths. If there are too many cars (workers) and not enough toll booths (managers), traffic stops. If there are too many toll booths, it's expensive and slow.
• The AI Solution: This new AI is like turning the highway into a magical, self-driving road. Suddenly, one toll booth can handle 20 cars instead of 3. The "coordination friction" disappears.
• The Result: The middle layer of managers (the Section Editors) becomes unnecessary. The Chief Editor can now talk directly to all the reporters. The company becomes flatter.

2. The Two Paths: The "Regime Fork"

The paper argues that this technology doesn't lead to just one future. It leads to a fork in the road, and which path we take depends on who gets the super-tool first.

Think of the AI tool as a Super-Backpack. It makes you stronger and lets you carry more weight.

Path A: The "Rising Tide" (The Good News)

• Scenario: Everyone gets a Super-Backpack. The Chief Editor has one, the Section Editors have one, and even the junior reporters get one to help them coordinate their own small projects.
• What happens: Because everyone is more efficient, the company can do more work. They start new projects (like a new podcast or a video channel) that were too hard to manage before.
• The Outcome: More people get jobs. Wages go up for everyone. The gap between the boss and the worker stays roughly the same. It's a "win-win."

Path B: The "Winner Takes All" (The Bad News)

• Scenario: Only the Chief Editor (the top 1% of managers) gets the Super-Backpack. The Section Editors and reporters get nothing.
• What happens: The Chief Editor becomes a superhero. They can now manage 100 people directly. The Section Editors are fired because they are no longer needed. The Chief Editor's team produces massive wealth, but they keep most of it for themselves.
• The Outcome: The company makes more money, but the gap between the Chief Editor and the workers explodes. The "superstar" gets richer, and the middle class of managers disappears.

3. The Big Surprise: The "Manager vs. Worker" Gap

The paper finds something interesting that happens in BOTH paths:

• The Gap Widens: Whether it's the "Good News" or "Bad News" path, the people doing the coordinating (the managers) always get richer relative to the people doing the work (the workers).
• Why? Because the AI makes the manager's job so much more powerful. They can oversee a giant army of workers. Since they are the "captain" of this giant ship, they claim a bigger share of the treasure.
• The Silver Lining: Even in the "Bad News" path, the paper finds that unemployment goes down. Because the company is so efficient, they can afford to hire more people than before. So, while the rich get richer, fewer people are left with no job at all.

4. The Real-World Lesson: It's About Access

The most important takeaway isn't about the technology itself; it's about who controls it.

• The Technology: AI that reduces coordination costs is inevitable. It will flatten hierarchies.
• The Variable: The paper introduces a concept called $\beta$ (Beta). Think of Beta as the "Access Key."
  • If Beta is Low: The key is cheap and available to everyone. We get the "Rising Tide."
  • If Beta is High: The key is expensive and only for the elite. We get the "Winner Takes All."

Summary Analogy: The "Conductor" vs. The "Orchestra"

Imagine an orchestra.

• Before AI: The conductor can only lead 20 musicians. To lead 100, you need 5 conductors.
• After AI: The conductor gets a "Magic Baton" that lets them lead 100 musicians directly.
• The Fork:
  • If every musician gets a mini-baton to help them stay in sync, the whole orchestra plays better, and everyone gets paid more.
  • If only the main conductor gets the Magic Baton, they become a god-like figure, the 5 middle conductors are fired, and the main conductor takes a huge cut of the ticket sales.

The Bottom Line:
AI isn't just about replacing workers; it's about reorganizing the workplace. Whether this leads to a fair, booming economy or a world of super-rich bosses and struggling workers depends entirely on policy. We need to make sure the "Magic Baton" (the AI coordination tools) is available to everyone, not just the people already at the top.

Technical Summary

Here is a detailed technical summary of the paper "AI as Coordination-Compressing Capital: Task Reallocation, Organizational Redesign, and the Regime Fork" by Alex Farach (February 2026).

1. Problem Statement

The dominant literature on AI and labor markets relies on task-based models (e.g., Acemoglu & Restrepo, 2019; Agrawal et al., 2026). These models treat organizational structure as exogenous (fixed), focusing solely on how AI changes the set of feasible tasks ($\bar{\tau}$) or worker capabilities ($\Omega(i)$).

• The Gap: Empirical evidence suggests AI significantly alters organizational architecture (flattening hierarchies, reducing management layers), yet current models cannot explain this because they do not model coordination costs endogenously.
• The Core Question: How does AI, by reducing the friction of coordinating human (or agent) labor, reshape organizational hierarchy, task creation, and income distribution? Does it lead to broad-based gains or "superstar" concentration?

2. Methodology and Model Framework

The author introduces a hierarchical production model where Agent Capital ($K_A$) is a distinct input that reduces coordination friction, rather than just augmenting labor or capital.

A. Production and Coordination

• Structure: A firm consists of managers ($i$) and workers ($j$). Team output follows a Cobb-Douglas form: $Y_i = A \cdot L_{eff,i}^\alpha \cdot m_i^{1-\alpha}$, where $m_i=1$ (one manager per team).
• Effective Labor ($L_{eff}$): Raw labor is discounted by coordination friction. For a team of size $n_i$ with total skill $Q_i$:
  $$L_{eff,i} = \frac{Q_i}{1 + c_i \cdot n_i}$$
  Here, $c_i$ is the coordination cost per worker. As $n_i$ grows, the denominator increases, creating diminishing returns due to coordination overhead.
• Agent Capital ($K_A$): Reduces coordination cost $c_i$ via the function:
  $$c_i(K_A) = \frac{c_0}{1 + \gamma \cdot K_A \cdot s_i^\beta}$$
  • $c_0$: Baseline friction.
  • $\gamma$: Effectiveness of AI in compressing coordination.
  • $s_i$: Managerial skill.
  • $\beta$ (Elite Complementarity): The critical parameter. It determines how $K_A$ interacts with skill.
    • If $\beta \approx 0$ (low): AI is general infrastructure, benefiting all managers equally.
    • If $\beta > 1$ (high): AI is elite complementarity, disproportionately amplifying the capacity of high-skill managers.

B. Key Mechanisms

1. Span of Control ($S_i$): Defined as the maximum team size a manager can handle: $S_i = 1/c_i(K_A)$. As $K_A$ rises, $c_i$ falls, and $S_i$ expands linearly.
2. Hierarchy Compression: As spans expand, the number of hierarchical layers required to manage a fixed workforce decreases: $Layers = \lceil \log(N)/\log(S) \rceil$.
3. Endogenous Task Creation: The feasible task frontier $T(K_A)$ expands with AI: $T(K_A) = T_0(1 + \delta \cdot K_A)$.
   • $\delta = 0$: Fixed task space (pure reallocation).
   • $\delta > 0$: New tasks are created, absorbing displaced labor.

C. Simulation Setup

The author simulates a "newsroom" economy with 20 heterogeneous managers and 400 workers.

• Matching: Positive Assortative Matching (PAM) assigns high-skill workers to high-skill managers.
• Regimes: The model sweeps across a $2 \times 2$parameter space defined by **$\beta$** (Low/High) and **$\delta$** (Low/High).

3. Key Contributions

1. Introduction of Agent Capital ($K_A$): Formalizes AI not as a task-substituting tool, but as a coordination-compressing capital that alters the fundamental constraints of organizational design (spans of control and hierarchy depth).
2. Five Formal Propositions:
   • P1 (Output): Output strictly increases with $K_A$ due to reduced friction.
   • P2 (Span Expansion): Managerial spans of control strictly expand.
   • P3 (Manager Demand): The total number of managers required strictly decreases (hierarchy flattening).
   • P4 (Wage Dispersion): Managerial wage dispersion increases with $K_A$, driven by $\beta$. High $\beta$ leads to "Winner-Takes-All" concentration; low $\beta$ leads to compression.
   • P5 (Task Frontier): If $\delta > 0$, the task frontier expands, increasing total employment.
3. The Regime Fork: The paper identifies a critical divergence where the same technology produces opposite distributional outcomes based on who controls the coordination gains (parameter $\beta$).

4. Results and Regime Analysis

The simulation reveals four distinct regimes based on the interaction of $\beta$ (access) and $\delta$ (task creation):

Regime	Parameters	Dynamics	Outcome
Gentle Compression	Low $\beta$, Low $\delta$	AI benefits all managers equally; no new tasks.	Moderate output growth; flat/slight inequality reduction.
Rising Tide	Low $\beta$, High $\delta$	Broad access + new tasks created.	Optimistic: High output, employment surges, inequality falls rapidly.
Winner Takes All	High $\beta$, Low $\delta$	AI amplifies elite managers; no new tasks.	Pessimistic: High output, but massive inequality; top managers absorb all gains.
Creative Destruction	High $\beta$, High $\delta$	Elite dominance + new tasks.	Volatile: High growth but high inequality; top talent dominates new sectors.

Critical Findings:

• Economy-Wide Inequality: Contrary to the manager-only view, the full-economy Gini coefficient falls in all regimes as $K_A$ rises. This is because coordination compression expands employment, moving workers from zero income (unemployed) to positive income.
• The Manager-Worker Gap: Despite falling overall inequality, the wage gap between managers and workers widens universally. Managers capture a fixed share of team output ($(1-\alpha)Y_i$), while the worker share ($\alpha Y_i$) is diluted across larger teams.
• The Role of $\beta$: The speed of inequality reduction depends entirely on $\beta$. Low $\beta$ (broad access) accelerates the "Rising Tide" effect; high $\beta$ (elite access) slows it, creating a "superstar" equilibrium.

5. Significance and Policy Implications

• Reframing AI Policy: The paper argues that the distributional impact of AI is not determined solely by the technology's capability ($\gamma$) or the nature of work ($\delta$), but by who controls the coordination elasticity ($\beta$).
• Three Policy Frames:
  1. Replacement Frame: Focuses on defensive redistribution (UBI) if AI displaces labor.
  2. Tool Frame: Focuses on upskilling workers to use AI.
  3. Coordination Frame (Proposed): Focuses on access and organizational design. Policy should ensure broad access to AI coordination platforms to keep $\beta$ low, preventing the "Winner Takes All" regime.
• Testable Predictions:
  • Firms adopting AI should show flattened hierarchies and reduced management layers (supported by recent empirical data from Ewens & Giroud, 2025).
  • Managerial wage dispersion should increase faster in industries where AI adoption is concentrated among elite decision-makers.
• Limitations: The model assumes a stylized Cobb-Douglas production function and fixed skill distributions. It abstracts from general equilibrium effects (firm entry/exit) and assumes coordination costs can theoretically approach zero (ignoring verification/audit floors).

Conclusion

The paper concludes that AI acts as a regime fork. It is a powerful engine for productivity and employment expansion, but its distributional consequences are bifurcated. If coordination-compressing AI is accessible as general infrastructure (low $\beta$), it democratizes gains. If it remains an elite complement (high $\beta$), it concentrates wealth at the top of the managerial hierarchy, even as overall unemployment falls. The critical policy lever is not just the technology itself, but the institutional arrangements governing access to it.