GO TO A SECTION

↑ ↓ to navigate · Enter to openExplore all sections →
FDEInterviews logo
RecentarXiv preprint

Organizing Intelligence

An economic model separates AI competence from coordination. Learn why stronger individual agents need not create stronger organizations, with a runnable example of the paper's equations.

Breakdown by FDE Interviews · 12 min read · Updated

About the original · 2026
AUTHORS
Imas, Koh, Legg
FROM
Google DeepMind; University of Chicago; Columbia University
PUBLISHED
Oct 2026 · arXiv preprint
IN THIS BREAKDOWN · 10 SECTIONS

You can read 14% of this breakdown.

01

The paper in one minute

A strong coding model is not automatically a strong engineering organization. Work can require several complementary skills, and the value of combining specialists depends on how well they coordinate. Organizing Intelligence puts that distinction into an economic model of firms containing humans, AI agents or both.

The authors separate competence, which changes productivity at a given organizational scale, from coordination, which changes how productivity behaves as the organization grows. Human expertise is limited in supply; AI expertise can be copied at a cost. Firms choose their composition and scale within the model's market environment.

One consequence is that an AI system can exceed average individual human capability while remaining less productive than a complementary human organization. Another is that improving AI competence can have different effects from improving coordination: one can change who does the work and favor smaller firms, while the other can support larger AI-heavy organizations under the specified assumptions.

This is theoretical research. Its numerical economy illustrates mechanisms; it does not estimate when jobs will disappear, predict an employer's staffing plan or measure the performance of a particular agent framework. The arXiv identifier begins with November, but the first submission was October 31, 2026.

THE IDEA, VISUALLY
TWO DIFFERENT WAYS TO IMPROVE AN ORGANIZATION Combine complementary skills Improve coordination Coder alone: (9, 1) Harmonic mean = 1.8 Equal coder + writer: (5, 5) Harmonic mean = 5 At coordination 0.75: best continuous size = 6 Higher competence raises output. Higher coordination changes how output scales with size. Section 2 numerical example and equations. These are model quantities, not staffing estimates. An individual benchmark does not measure the capability a complementary team can assemble.
The paper's illustrative economy separates competence from coordination. A coder with skills (9,1) has harmonic-mean competence 1.8; an equal coder-writer team averages (5,5), giving competence 5. The scale formula produces a continuous optimum of 6 at coordination 0.75. These are model quantities, not measurements of real teams.
Research notes: the problem, approach and results
1.8 → 5

standalone specialist versus balanced-team competence in the section 2 example

6

best continuous firm size at coordination 0.75 in the illustrative function (section 2.1)

2 effects

technology and reorganization components of wage changes (section 6.3)

THE PROBLEM

Individual capability benchmarks do not explain how complementary workers organize, how coordination limits scale, or how firms respond to reproducible AI labor.

THE IDEA

Model competence and coordination separately, then let humans and AI sort into competing firms under specified production and market assumptions.

THE RESULT

The model permits smaller firms after competence gains and larger concentrated firms after sufficient coordination gains. Employment and wage effects depend on reorganization.

02

Why it matters to an FDE

A customer may ask how many employees a new model can replace after seeing a coding benchmark. That question skips the organization around the task. Someone supplies requirements, resolves conflicting changes, validates outputs and handles exceptions. A model can improve at one step while the entire workflow remains limited elsewhere.

Use this paper to structure a deployment investigation. Measure individual task quality and the cost of coordinating the combined workflow separately. If more agents create merge conflicts or duplicate investigations, adding model capacity may increase spend without improving accepted output.

The paper also cautions against treating human value as a fixed ranking from a standalone test. A person with modest isolated task output can contribute by helping specialists work together. That does not prove a particular role is protected from automation; it identifies a contribution that a solo benchmark can miss. Our multi-agent orchestration concept connects this organizational question to the mechanics of an agent system.

THE FULL BREAKDOWN

Understand the mechanism. Try it on a real problem.

Unlock this breakdown and 294 more papers.

Premium opens every paper in full, plus every answer, concept, course and guide on the site. One payment for six months of access.

6 months · One payment · No auto-renewal

Get Premium · ₹2,000 / $25
  • ◇1 architecture diagram
  • </>2 runnable code blocks
  • ▦3 results tables
  1. 03Key takeaways

    Distinguish complementarity, coordination and equilibrium responses before making a forecast

    Free account
  2. 04How it works

    Derive the competence and scale mechanisms from the paper's illustrative economy

    Premium
  3. 05The numbers that matter

    Treat numerical parameters as an illustrative economy, not empirical labor estimates

    Premium
  4. 06Try it yourself

    Reproduce the competence arithmetic and see how coordination moves the optimum

    Premium
  5. 07Where it breaks

    Identify the assumptions that would need evidence before using the model for a forecast

    Premium
  6. 08Using it on a deployment

    Design separate tests for task competence and coordination costs

    Premium
  7. 09Interview angles

    Explain why a model benchmark cannot determine the value of a human-AI organization

    Premium
Back to the paper library →