---
title: "Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.25446v1"
permalink: "https://feed.astralor.com/items/itm_8409a158dfbfc610ae21ffa4"
published: "2026-07-28T08:35:21.000Z"
score: 81
channel: "ai"
tags: ["IMACS","LLM","Contextual Bandits","Multi-agent Systems","Organizational Theory","Adaptive Routing"]
---

# Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm

## AI commentary

> This paper brings real organizational theory from management science into multi-agent systems, not just as an analogy but as executable, verifiable configurations. For engineers and researchers working on agent orchestration, it offers both principled insights and practical recipes—especially the finding that organizational design is not one-size-fits-all, but must be recalibrated for each model family.

## At a glance

Multi-agent frameworks often entangle organization, coordination, and collaboration protocol. This paper proposes IMACS, which decouples these three into orthogonal, swappable layers, making classic organizational theory (Belbin roles, Mintzberg coordination, RACI accountability) an executable configuration, and unifies six published collaboration algorithms behind a common interface. It further introduces Adaptive Org Routing, a contextual-bandit meta-protocol that selects a protocol per task under an explicit quality-cost tradeoff, outperforming every fixed protocol in a controlled study. Ablations reveal that accountability placement effects depend on whether the protocol routes the deliverable through the accountable agent, and the winning placement flips across model families, proving that organizational design must be revalidated or learned for each model binding.

## Key points

- Proposes IMACS framework decoupling organization (Who), coordination (How), and collaboration protocol (Which Algorithm) in multi-agent systems.
- Maps classic organizational theories (Belbin roles, Mintzberg coordination, RACI accountability) into executable configurations.
- Adaptive Org Routing, a contextual-bandit meta-protocol, dynamically selects the best collaboration protocol per task.
- Experiments show accountability placement effects depend on protocol routing, and optimal design varies across model families, forbidding hard-coding.

## Full text

Multi-agent frameworks built on large language models (LLMs) routinely entangle three logically distinct concerns: who is on the team (organization), how members align (coordination), and which algorithm fuses their work (collaboration protocol). IMACS (Intelligent Multi-Agent Collaboration System) separates the three into orthogonal, independently swappable layers. Classic organizational theory (Belbin roles, Mintzberg coordination, RACI accountability) becomes executable, validated configuration, and the framework places six published collaboration algorithms behind a common interface while exposing roles, coordination, and accountability as independently configurable factors. We use this separation to conduct controlled comparisons in which organizational assignments vary while the collaboration protocol is held fixed. It also turns protocol choice into a variable that can be learned: Adaptive Org Routing, a contextual-bandit meta-protocol, selects a protocol per task under an explicit quality-cost tradeoff, outperforms every fixed protocol in a controlled study, and trains online on real benchmark and LLM-judge rewards. The ablations expose a mechanism. Accountability placement changes outcomes exactly when the protocol routes the deliverable through the accountable agent, and the winning placement flips across model families, so organizational design cannot be hard-coded; it must be revalidated, or learned, for each model binding.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_8409a158dfbfc610ae21ffa4