One idea explains ant colonies, companies, and software: coordination is just managing what depends on what.
What makes coordination theory work across disciplines like biology and computer science?
This explores why one coordination framework can describe ant colonies, companies, and computer programs alike, and what the collection shows about where that cross-domain transfer holds up and where it breaks down.
This explores why one coordination framework can describe ant colonies, companies, and computer programs alike, and where that transfer holds up. The short answer from the corpus: coordination theory travels well because it ignores *who* is coordinating and asks only *what depends on what*. Malone and Crowston's 1994 definition, coordination as managing dependencies between activities, is deliberately substrate-neutral Can we match dependency types to coordination processes?. Two tasks might share a resource, one might need the other's output, or both might have to happen at the same moment. Those dependency types look the same whether the actors are neurons, warehouse workers, or software processes. Once you name the type, you can match it to a known fix, such as a queue, a handoff, or a schedule.
The theory also travels because its trade-offs come from structure, not material. Malone and Crowston's formal models show that centralized control is cheap to coordinate but fragile when the central node fails. Decentralized markets survive failures but spend heavily on coordination. Hierarchies sit in between and leave spare capacity unused How do coordination architectures trade off efficiency against resilience?. A biologist comparing a queen-led hive to a leaderless swarm and an engineer choosing between an orchestrator agent and peer-to-peer agents are working on the same problem. The same trade-offs show up in current LLM agent networks: coordination breaks down predictably as the network grows, either because agents agree too late or because they adopt a plan without telling their neighbors Why do multi-agent systems fail to coordinate at scale?.
The most literal biology-to-computing bridge is stigmergy, the way termites coordinate by leaving traces in their environment that shape what others do next, with no direct communication. AI agents that coordinate through shared storage look a lot like this, but the corpus is honest that the link is still loose. Nobody has spelled out which properties of stigmergy actually carry over Does storage-mediated coordination work like stigmergy?. Similar examples appear under other names. MetaGPT's agents coordinate better by posting standardized documents to a shared workspace than by chatting Does structured artifact sharing outperform conversational coordination?. Reasoning models that share a working memory (a concurrent KV cache) start dividing up work and spotting redundancy without any training to do so Can multiple LLMs coordinate without explicit collaboration rules?. In both cases the shared environment does the coordinating, which is the core idea of stigmergy.
A second reason the theory transfers is that it splits into layers you can swap independently. MOISE+ separates who holds which role, what the overall plan is, and who is permitted or obliged to do what Can separating structure and function improve multiagent organization design?. IMACS goes further and borrows from management science: Belbin's team roles for structure and Mintzberg's coordination types for mechanism, applied to AI agent teams as interchangeable modules Can multi-agent teams swap roles, coordination, and algorithms independently?. Organizational theory written for humans can be plugged into software almost directly, because it was already abstract.
One caveat is easy to miss. Agents that behave alike are not necessarily coordinating. They may just be responding to the same outside cause, so telling the two apart takes evidence that influence actually passed between them, or a test that cuts the communication channel and checks whether the shared behavior survives How do we tell coordination apart from shared causes?. This is a standard problem in biology, where synchronized animals can be following each other or just reacting to the same weather, and it matters just as much for AI systems. The collection is strong on the computing side and on organizational theory, but thin on biology itself. Stigmergy is the main biological thread, and it is acknowledged as underdeveloped.
Sources 9 notes
Malone and Crowston's 1994 framework defines coordination as managing dependencies between activities and proposes that identifying dependency types enables systematic matching to coordination processes. The definition bridges multiple disciplines by treating coordination as a universal mechanism across human, computational, and biological systems.
Malone and Crowston's formal models show centralized schemes minimize coordination costs but fail under processor failures, decentralized markets resist failures but incur high coordination costs, and hierarchies reduce coordination costs while creating unused capacity. Each architecture optimizes one dimension at the expense of others.
AgentsNet benchmark shows agents fail to coordinate strategies either by agreeing too late or adopting strategies without informing neighbors. Agents accept neighbor information without verification, enabling error propagation while remaining capable of detecting direct conflicts.
The paper connects agent coordination through shared storage to stigmergy, where traces in a medium drive later action. However, the excerpt does not define stigmergy as used, specify which properties transfer, or explain the defensive value beyond calling it a shared store.
MetaGPT demonstrates that agents producing standardized engineering documents achieve superior coordination compared to conversational exchange. Active information pulling from shared environments eliminates noise and mirrors efficient human workplace infrastructure.
Show all 9 sources
Existing reasoning-capable models like QwQ and DeepSeek-R1 spontaneously formulate plans, detect redundancy, and adapt strategies when given shared access to a concurrent KV cache. This coordination emerges without fine-tuning, suggesting reasoning models already possess multi-agent collaboration capabilities.
Hübner, Sichman, and Boissier's 2002 model splits organizational specification into structural (roles and groups), functional (global plans), and deontic (permissions and obligations) dimensions. This separation avoids failures where agents lack guidance or lose validated plans.
IMACS demonstrates that team structure (Belbin roles), coordination mechanism (Mintzberg framework), and fusion algorithm can be architecturally separated into independent components. This decoupling lets teams and protocols swap without rewriting each other, unlike existing frameworks that fuse these concerns together.
Similar behavior between agents has two sources: direct influence or shared external causes. Distinguishing them requires either observational evidence of transfer or interventional tests like closing channels to see if behavior recurs.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Towards a Science of Scaling Agent Systems
- Drop the Hierarchy and Roles: How Self-Organizing LLM Agents Outperform Designed Structures
- Worse Together: How Performance Breaks Down in Multi-User Multi-Agent Teams
- Toward an Organizational Science of Multi-Agent LLM Systems: Decoupling Who, How, and Which Algorithm
- AgentsNet: Coordination and Collaborative Reasoning in Multi-Agent LLMs
- Embodied LLM Agents Learn to Cooperate in Organized Teams
- The Interdisciplinary Study of Coordination
- Comparing Human Oversight Strategies for Computer-Use Agents