Daemon Corporation

Global Leader in Excellence.

Research Report № 007

Systems for Coordinated Action: A Vehicle for Comparison, Not a Domain

Central finding

“Systems for coordinated action” is not a coherent substantive domain of inquiry. The proposed umbrella combines problem families with different explananda, causal variables, success criteria, normative commitments, and standards of evidence. Synchronizing distributed processes, overcoming free-riding, making legitimate public decisions, mobilizing a movement, administering a firm, constituting a group agent, revising institutions, and orchestrating AI agents are not variants of one underlying process.

But the umbrella is not incoherent either. A thin and useful common object survives:

A coordination arrangement is a configuration of mechanisms, interfaces, roles, rules, signals, or mutual adjustments that shapes two or more distinguishable activities whose feasible execution or consequences are outcome-relevantly interdependent.

The corresponding inquiry is best called coordination of interdependent activity. Its comparative object is the coordination arrangement, not a presumptively unified kind of system.

This frame can compare markets, firms, commons, protocols, crowds, movements, public institutions, interorganizational networks, human–AI teams, and multi-agent AI systems. Its sharpest exclusion rule is:

No outcome-relevant dependency, no coordination.

That rule excludes merely parallel and correlated behavior while admitting unintended, emergent, coercive, technical, and failed coordination. Yet the price of this reach is substantial. The definition does not itself explain intentions, incentives, legitimacy, power, distribution, or collective agency. Those must be restored as typed, domain-specific variables rather than treated as universal functions.

The best precedent is the scholarship on “governance.” After surveying nine disciplinary literatures, Kees van Kersbergen and Frans van Waarden concluded that governance was too theoretically and methodologically diverse to become a shared theory, but could become “a vehicle for comparison and for mutual learning and theoretical inspiration.”1 The same verdict fits here.

The practical recommendation is therefore:

  1. Retire “systems for coordinated action” as a field name.
  2. Use coordination of interdependent activity as the problem label.
  3. Use coordination arrangement as the success-neutral unit of comparison.
  4. Treat the common vocabulary as a translation discipline, not a predictive design theory.
  5. Replace the proposed nine universal functions with a five-part operational kernel plus optional social and institutional modules.
  6. Analyze power and contestation alongside the coordination core whenever human interests or asymmetric authority are involved.

1. The underlying problem: one word conceals two boundaries

The broadest established cross-disciplinary definition comes from Thomas Malone and Kevin Crowston: coordination is the management of dependencies between activities. Their test is deliberately minimal: “if there is no interdependence, there is nothing to coordinate.”2 This definition travels well because it says little about the nature of the activities or actors. It can describe tasks in a firm, prices in a market, messages in a distributed system, or even the muscles of one body.

That portability also prevents it from individuating a domain. Malone and Crowston acknowledge that there is no single correct way to decompose a situation into actors and activities. Coordination can therefore be found at multiple scales, depending partly on the observer’s chosen system boundary. Francis Heylighen pushes the structural approach further, describing stigmergic coordination without planning, communication, mutual awareness, commitment, centralized control, or even, in the limiting case, agents.3

A second tradition draws the boundary elsewhere. Philosophies of shared and collective action ask whether participants are intentionally doing something together. Matching individual intentions are insufficient: two people independently intending to visit the Taj Mahal tomorrow do not thereby intend to visit it together. In Searle’s canonical contrast, dancers converging in a park count as acting together, while picnickers independently running toward the same shelter do not; the difference is that the dancers’ convergence is something at which the participants aim.4

These are not two formulations of the same boundary:

  • The structural criterion asks whether activities are causally, informationally, institutionally, or anticipatorily coupled.
  • The intentional criterion asks whether the joint pattern is part of what participants intend.

They cross-classify markets, stigmergy, crowd synchrony, and many technical systems. Pedestrians who inadvertently synchronize through the motion of a bridge are coordinated under the structural criterion because the bridge couples their movements, but they are not engaged in shared action under the intentional criterion.5

No inspected source resolves this dispute. The broad comparative framework must therefore make a declared choice rather than claiming to discover a settled meaning. This report adopts the structural criterion, because otherwise markets, technical distributed systems, stigmergic processes, and much of multi-agent AI disappear from the comparison. That is a stipulation, not a finding about ordinary language.

Working definitions

The following definitions are designed to keep adjacent concepts distinct.

Concept Precise working account Necessary features
Coordination Management, or attempted management, of outcome-relevant dependencies among distinguishable activities Multiple distinguishable activities; dependency; coupling or adjustment
Coordination arrangement A persistent configuration of mechanisms, interfaces, roles, rules, signals, or mutual adjustments shaping interdependent activities The coordination core plus a recognizable configuration
Coordinated action Action by multiple action-capable units whose contributions are functionally coupled with respect to an activity, object, constraint, or outcome Action-capable units; functional coupling
Cooperation A strategic or motivational condition in which actors contribute toward compatible benefits despite opportunities or incentives not to contribute A contribution problem; opportunity for defection or opportunism
Collective action Mobilized contribution by multiple actors toward a collective good, claim, or outcome under participation, resource, free-riding, or opposition constraints Collective outcome or claim; contribution problem
Joint or shared action Intentional acting together through interlocking intentions or mutually responsive plans Joint intentional structure
Collective agency The capacity of a group to pursue purposes flexibly, authorize and construct a group voice, and undertake commitments across situations Organizational procedures supporting purposive flexibility, voice, and commitment
Organization A relatively durable arrangement of differentiated roles, routines, authority, information processing, and coordinated activity Durability and role differentiation
Governance Processes and rules that establish, authorize, constrain, and revise collective choices and relationships among actors Standing or authority; rules for collective choice
Institutional design Deliberate alteration of rules, positions, information, authority, incentives, and evaluation arrangements structuring an action situation Intentional intervention in a rule-structured situation

These definitions are stipulative because the underlying usage is not settled. A systematic review covering nine major management journals from 1948 through 2017 found that only 11 percent of articles using “coordination,” “cooperation,” or “collaboration” defined the term they used.6 At least three incompatible conventions coexist:

  1. Coordination is the genus, with cooperation, competition, and political processes treated as ways of managing dependencies.
  2. Coordination addresses informational or action-alignment problems, while cooperation addresses conflicting interests and incentives.
  3. Coordination is joint deliberation over common goals, while cooperation is implementation of those goals.

For comparative work, the second convention is most useful because it keeps operational coupling and incentive alignment analytically separate. It should nevertheless be announced whenever used.

“Collective agency” is itself ambiguous

Three distinct ideas are often placed under this heading:

  • Group agency: the group has procedures for forming and acting on attitudes attributable to it.
  • Shared agency: individuals intentionally act together through interlocking plans.
  • Collective efficacy: members share a belief in their joint capability.

These differ in bearer, truth conditions, and empirical method. The first concerns an organized group; the second concerns a relationship among individual intentions; the third is a psychometric belief construct. They should not be treated as one layer of a unified domain.7

Philip Pettit’s account supplies the most discriminating sense for present purposes: group agency requires purposive flexibility, individual acquiescence, authorization and construction of a group voice, and capacity for commitment. An ad hoc group can jointly rescue someone without acquiring the procedures necessary to act as an agent across varied situations.8 Collective agency is therefore an achieved subset of organized action, not a synonym for coordination.

2. What is necessary—and what is merely contingent

Under the structural criterion, only three features are necessary:

  1. Two or more distinguishable activities or contribution streams
  2. An outcome-relevant dependency between them
  3. A coupling, adjustment, or control relation that manages or attempts to manage that dependency

The following are contingent:

  • shared goals;
  • cooperation or benevolence;
  • voluntariness or consent;
  • planning;
  • communication;
  • mutual awareness;
  • simultaneous presence;
  • commitment;
  • centralized supervision;
  • authority;
  • formal rules;
  • legitimacy;
  • learning;
  • collective intention;
  • collective agency;
  • successful performance.

Stigmergy supplies the most systematic counterexample set: coordination can occur through environmental traces without direct communication, mutual awareness, imposed sequencing, centralized supervision, or durable commitment.3 Markets supply another. Prices couple dispersed decisions even though participants neither survey the whole field nor jointly deliberate about a collective objective.9

Coercion is also not an extensional boundary. Pettit’s demanding account of group agency permits member acquiescence that is intentional but neither discretionary nor willing, using employment under the threat of starvation as an example.8 If voluntariness is not required even at the group-agency end of the spectrum, it cannot be required for the thinner concept of operational coordination. A conscript army, coercive workplace, or algorithmically managed labor platform can therefore be coordinated without being cooperative, consensual, or legitimate.

That conclusion must not be misread as normative neutrality. A thin coordination description can say that an algorithmic scheduling system manages task and resource dependencies. It cannot, by itself, explain who controls the system, whose interests it serves, who can exit, or why workers resist it. Research on algorithmic control describes six mechanisms—restricting, recommending, recording, rating, replacing, and rewarding—and shows how they generate organized counter-action or “algoactivism.”10 The coordination account and the power account describe the same arrangement, but only the latter explains its contested development.

Attempt, arrangement, and outcome

A success-neutral inquiry must distinguish:

  • Coordination attempt: a mechanism is being applied to a dependency.
  • Coordination arrangement: the configuration through which coupling is attempted.
  • Coordinated outcome: the dependency was actually managed to the relevant standard.

Without this distinction, “coordination” silently shifts between a process and a successful result. Failed protocols, ineffective governance arrangements, and poorly orchestrated teams remain within the comparative frame even when they do not produce coordinated outcomes.

3. The smallest coherent map

The eight proposed problem classes are best treated as partially overlapping problem families. Operational dependency management is the thin common core. Cooperation, governance, organization, and collective agency add variables that are not reducible to that core. Adaptation cuts across the families rather than forming a higher layer.

Problem family Central problem Typical failure Relation to the core
Dependency management and synchronization Shared resources, prerequisites, simultaneity, causal order Collision, duplication, deadlock, inconsistent ordering Minimal core
Cooperation and incentive alignment Joint benefit with temptation to defect or shirk Free-riding, opportunism, adverse selection Adjacent strategic layer
Collective choice and governance Aggregating preferences, values, authority, and jurisdiction Cycles, domination, illegitimacy, noncompliance Partially distinct normative domain
Mobilization and collective action Recruiting contributions under resource, opposition, and participation constraints Apathy, fragmentation, demobilization, co-optation Specialized contribution problem
Organization and administration Maintaining differentiated work through roles, routines, authority, and information processing Overload, silos, goal displacement, coordination cost Durable implementation family
Collective agency Forming and enacting attitudes attributable to a group Incoherence, unauthorized voice, inability to commit Strong achieved subset
Institutional adaptation and learning Revising rules and routines under delayed or ambiguous feedback Lock-in, success traps, maladaptation Cross-cutting dynamic
Adaptive multi-actor systems Coordinating under partial observability, communication limits, learning, and nonstationarity Cascading error, semantic drift, overhead, unsafe emergence Technical implementation family spanning several layers

They are not one causal domain because their dependent variables differ. A distributed-systems proof establishes possibility or impossibility under a formal timing and failure model. A trauma-center ethnography identifies situated practices under urgency and interpretive disagreement. A governance study evaluates legitimacy, trust, authority, and compliance. A movement study asks why participation or mobilization varies. A multi-agent AI benchmark measures task performance under a specified architecture.

Even similarly named limitations cannot be directly transported. The Fischer–Lynch–Paterson result concerns deterministic consensus in asynchronous message-passing systems with one crash failure.11 Arrow’s theorem concerns aggregation of ordinal preferences under a specified conjunction of rationality and fairness conditions.12 The discursive dilemma concerns the relation between group-level consistency and responsiveness to members’ judgments. These are not instances of one general “impossibility of collective decision” unless their primitives are explicitly mapped.

A non-nested taxonomy

A useful taxonomy is therefore a branching map, not a ladder on which every case climbs toward full collective agency.

Position Characteristic question Status
0. Shared exposure Are units merely responding to the same external condition? Excluded unless coupling is present
1. Operational coordination How are timing, sequencing, allocation, or compatibility dependencies managed? Minimal coherent core
2. Strategic cooperation Why contribute rather than defect? Adds incentives and payoff structure
3. Joint action Are participants intentionally doing something together? Adds interlocking intentions
4. Governance Who may decide, under what rules and standards of legitimacy? Adds constitutive and normative structure
5. Organization How is differentiated action maintained over time? Adds roles, administration, and continuity
6. Collective agency Can a group form a voice and undertake commitments across cases? Achieved side branch, not universal culmination
Cross-cutting: adaptation How do mechanisms and rules change with feedback? Possible at every level; neither necessary nor invariably beneficial

Governance is not simply operational coordination plus more rules. Jurisdiction, public justification, and legitimacy are different kinds of variables. Collective agency likewise is not the top rung of every coordinated system: a market can coordinate without becoming an agent, and a tightly coupled technical protocol does not acquire a group voice.

4. Boundary stress tests

The structural criterion separates coordination from mere resemblance by asking whether one activity’s information, resource use, timing, output, or anticipated behavior enters the conditions under which another is selected or succeeds.

Case Classification Reason
People independently opening umbrellas in rain Excluded Shared cause, but no coupling among their activities
Coincident outputs from uncoupled generators Excluded Correlation without causal, informational, institutional, or anticipatory dependence
Two populations exposed to the same price shock Excluded unless their responses interact Common exposure alone is not coordination; market clearing may create coupling
Logical clocks and synchronization protocols Included Ordering and simultaneity dependencies are explicit; legitimacy and motivation are irrelevant13
Unintended crowd synchrony Included under the structural criterion; excluded under the intentional criterion A shared medium couples behavior, although no participant aims at the collective pattern5
Coercive command Included Authority can align interdependent activities without cooperation, consent, or shared purpose
Unilateral control of passive components Normally excluded Control is not multi-actor coordination unless the components retain analytically relevant decision or failure autonomy
Markets Included Prices and market-clearing processes couple decentralized choices without collective intention9
Crowds Included when reciprocal adjustment or a shared medium couples trajectories Aggregate pattern alone is insufficient; collision avoidance or phase coupling can qualify
Social movements Included Mobilization, resource aggregation, opposition, and contribution thresholds create interdependencies; organizational unity is contingent
Firms Included Roles, authority, routines, and information systems manage differentiated work
Public institutions Included Operational coordination combines with authoritative collective choice, administration, and legitimacy
Commons Included Rules, boundaries, monitoring, sanctions, and nested arrangements manage provision and appropriation14
Interorganizational networks Included Coordination and governance occur without unitary hierarchy; effectiveness is contingent on trust, size, goal consensus, and task characteristics15
Emergency response Included Usually combines command, interagency coordination, temporary organization, and local adaptation; no single “hierarchy” or “network” label is sufficient16
Technical distributed systems Included Causal ordering, agreement, synchronization, and fault handling suffice
Communities Conditional Community identity or co-presence is insufficient; inclusion requires rules or interactions managing an actual dependency
Open-source ecosystems Included Modular contribution still requires integration, quality control, interfaces, and project governance17
Human–AI systems Conditional Mere tool use is insufficient; inclusion requires interdependent roles, communication, adjustment, or shared task state
Multi-agent AI systems Included Roles, communication topology, synchronization, memory, aggregation, and termination are engineered coordination variables

Two difficult edges remain.

First, emergent order may be causally coupled without being usefully described as coordination. The structural criterion deliberately includes such cases, but “collective dynamics” may sometimes be the clearer label.

Second, adversarial interdependence is unresolved. Competitors and enemies adjust to one another through outcome-relevant dependencies. Including them risks expanding coordination toward strategic interaction generally; excluding them makes compatible outcomes an undeclared necessity.

5. What transfers—and what does not

The strongest positive case for the umbrella is not a universal mechanism. It is an ordering of what travels well:

  1. Problem statements and dependency types transfer most readily.
  2. Mechanism families, interfaces, and compositions transfer conditionally.
  3. Concrete implementations and evidence usually do not transfer.

The following matrix summarizes that judgment. It is a disciplined analytical hypothesis, not the result of a direct eight-way comparative experiment.

Codes: C = relatively clean transfer; M = transfers with changed meaning or operation; K = conditional on explicit assumptions; B = generally breaks down.

Columns: DS = dependency/synchronization; CI = cooperation/incentives; GV = governance; MB = mobilization; OA = organization/administration; CA = collective agency; AL = adaptation/learning; AM = adaptive multi-actor systems.

Transfer object DS CI GV MB OA CA AL AM
Problem statement C C C C C C C C
Dependency type C C M M C M M C
Mechanism family C K K K C K K C
Enabling conditions C M M M M M C M
Interface abstraction C M M M C M M C
Composition and modularity C K K C C K C C
Concrete implementation K B B B K B K K
Evidence or effect estimate K B B B K B K K

The top row is clean only when functions are restated as neutral problems—“order dependent activities,” “allocate a scarce resource,” or “form an authoritative choice.” It does not mean that a socially loaded function such as legitimacy or commitment transfers intact.

Semantic erasure

Cross-domain language can appear to transfer while the construct changes identity:

  • accountability → logging
  • commitment → reward shaping
  • deliberation → message exchange
  • legitimacy → authorization
  • shared understanding → compatible state representation

These translations can be useful engineering analogies. They do not preserve answerability, obligation, public justification, consent, or mutual recognition. No robust evidence establishes measurement equivalence between these human institutional constructs and their technical counterparts.

Evidence on actual transfer

The mirroring hypothesis—that organizational relationships tend to track technical dependencies—is the best-measured transfer proposition in the evidence base. A review of 142 empirical studies found strong support in 70 percent of descriptive firm and industry studies, partial support in 22 percent, and no support in 8 percent. In open collaborative projects, the pattern reversed: 56 percent did not support mirroring. “Actionable transparency” in digital artifacts may allow broad visibility and rapid testing to substitute for formal organizational ties.18

That is evidence for conditional transfer, not a universal law. The same review observes that temporary contributor groups may develop dense communication ties while working, which partly re-scopes rather than decisively falsifies mirroring.

More serious limits come from two other sources:

  • Elinor Ostrom reported trying to identify rules that worked across ecological, social, and economic settings, finding that the rules associated with success and failure varied extensively, and eventually abandoning rule-level transfer in favor of more abstract “core underlying lessons.” She later suggested that “best practices” would have been a less misleading term than “design principles.”14
  • Samer Faraj and Yan Xiao’s trauma-center study argues that practices needed under urgency, novelty, surprise, and contested interpretation are highly situated and cannot be prespecified like traditional coordination mechanisms. It explicitly describes the dependency-typology approach as potentially too formal for such settings.19

Kevin Crowston, while developing a taxonomy of dependencies and mechanisms, likewise observed that earlier descriptions were too general to determine which alternatives would be useful in a given circumstance.20 The recurrent grammar is therefore descriptively plausible but not demonstrated to be predictively valid.

Failure-sampled evidence

The broad literature is weighted toward persistent or exemplary arrangements. One study that sampled failure directly examined 104 deprecated open-source projects and obtained maintainer-attributed explanations for 110 cases. Only three were attributed to conflicts among developers, the one category clearly identifiable as a coordination or governance failure. More common explanations included replacement by a competitor, obsolescence, lack of time, and lack of interest.21

This single study does not establish that coordination is generally unimportant to organizational mortality. It does show that a coordination vocabulary can describe the structure of an activity without explaining why the activity survives or dissolves. The distinction between descriptive reach and causal purchase is essential.

AI as a revealing transfer test

Recent multi-agent AI evidence is two-sided.

An experiment instantiating several historical governance templates in LLM collectives found best-to-worst performance gaps of roughly 57–61 percentage points on a single-turn tool-use benchmark. On a harder multi-step reasoning benchmark, the gaps narrowed to 8.2–11.8 points.22 A separate paired noise-floor study reported a 12-point upper confidence bound for one local clean contrast, while emphasizing that its comparison was heterogeneous and that the floor varies by model and domain.23 It cannot be treated as a universal threshold. Taken together, the results suggest that governance topology can be consequential when task difficulty does not dominate, while smaller effects on harder tasks remain uncertain.

A multi-agent failure taxonomy further reports 14 failure modes across three categories, but its headline category proportions were generated by an LLM judge over 1,642 traces, validated against human experts with κ = 0.77. The taxonomy itself originated in expert examination of more than 150 traces, while the reported κ = 0.88 referred to a 15-trace annotation exercise.24 Those scopes should not be conflated.

Human–AI teamwork shows similar limits. Reviews identify role ambiguity—whether an AI advises, executes, or shares responsibility—and emphasize calibrated rather than simply increased trust. One reported case found that adding an AI teammate reduced performance because coordination costs exceeded the benefits.25 Treating an AI tool as a teammate is therefore a design choice, not proof of symmetric agency, responsibility, or standing.

The appropriate portable artifact: an assumption ledger

A cross-domain claim should not move without recording at least:

  1. Actor model: humans, organizations, processes, software agents, or other units
  2. Objective structure: whose outcome criterion is being applied
  3. Information distribution
  4. Communication model
  5. Authority and power relations
  6. Temporal constraints
  7. Failure model
  8. Strategic behavior: whether defection or manipulation is possible
  9. Normative criteria: legitimacy, consent, fairness, and distribution
  10. Unit of evidence: proof, experiment, benchmark, case study, ethnography, or comparative dataset

The common contribution is thus not a portable recipe. It is a way to make otherwise incommensurable claims explicit enough to compare.

6. Why the nine proposed functions are not universal

The proposed functions are:

  1. constitute and legitimize;
  2. orient and frame;
  3. sense and interpret;
  4. deliberate and decide;
  5. mobilize and allocate;
  6. commit and motivate;
  7. coordinate execution;
  8. monitor and account;
  9. learn and adapt.

As a checklist for organizations and public institutions, this is useful. As a universal model of coordination, it fails.

Proposed function Assessment
Constitute and legitimize Institutional module. Essential in some governance settings; absent from logical clocks, markets, stigmergy, and many technical protocols.
Orient and frame Conditional. Becomes objective specification or state representation in technical systems, losing the interpretive sense of framing.
Sense and interpret Near-general only after translation. Sensing state is not always interpretation.
Deliberate and decide Governance and agency module. Algorithmic aggregation or consensus should not automatically be called deliberation.
Mobilize and allocate Must be split. Allocation is widespread; mobilization presupposes a contribution or participation problem.
Commit and motivate Strategic and social module. Rewards or persistence mechanisms in software are not equivalent to human commitment.
Coordinate execution The only near-universal function.
Monitor and account Must be split. Observability and logging travel broadly; answerability and sanction do not.
Learn and adapt Cross-cutting but not necessary. Fixed protocols and routines coordinate; adaptation can also produce lock-in or exploitation traps.26

At least four functions—constitute and legitimize, deliberate and decide, commit and motivate, and monitor and account—have no non-metaphorical application across all required cases. Markets allocate without collective deliberation. Stigmergic processes coordinate without commitment. Technical systems log events but do not thereby become accountable in the institutional sense.

Replacement: a coordination kernel plus optional modules

A more defensible universal kernel is:

  1. Delimit the activities, dependencies, and relevant system boundary.
  2. Represent or expose state, capabilities, constraints, and interfaces.
  3. Select or instantiate a coupling mechanism.
  4. Execute by allocating, ordering, synchronizing, or integrating contributions.
  5. Detect, contain, and repair coordination breakdowns.

Optional social and institutional modules are added only where their conditions exist:

  • constitute membership and standing;
  • legitimize authority;
  • construct shared orientation;
  • align incentives and commitments;
  • form collective choices;
  • assign accountability and sanctions;
  • learn and revise rules.

The five-part kernel is an evidence-informed synthesis, not a validated canonical model. Its advantage is negative: it avoids pretending that governance, motivation, agency, or adaptation is present in every coordination arrangement.

7. Comparison with rival framings

Candidate What it captures Why it is not the general answer
Systems for coordinated action Designed supports for collective performance “For” implies purpose, “coordinated” implies success, and “action” is ambiguous across social and technical systems
Collective agency Group voice, attitudes, commitments, and purposive action Too narrow for markets, crowds, protocols, and most temporary collectives
Collective action systems Mobilization, public goods, commons, and contribution problems Weak for synchronization, markets, and routine administration
Coordination systems Broad operational and engineering problems Often reads as a computing or control term and can obscure power and legitimacy
Coordination architecture Roles, interfaces, topology, and designed mechanisms Too design-centered for markets, emergence, crowds, and movements
Institutional design Rules, authority, incentives, and governance Excludes emergent and noninstitutional coordination; “design” overstates intentionality
Organized action Social action structured by interdependence and power Strong social explanation, but weak for agentless, market, and protocol cases
Governance Authority, rules, legitimacy, accountability, and collective choice Misleading for clocks, queues, and pedestrian synchrony
Social organization Broad social structure Excludes technical systems and remains too diffuse for precise mechanism comparison
Organizational design Durable roles, task structures, and administrative mechanisms Biased toward bounded organizations
Adaptive institutional systems Institutional learning and rule revision Excludes fixed and noninstitutional arrangements
Interdependent action systems Interdependence and failure as well as success Risks including every adversarial or unmanaged interaction
Coordination of interdependent activity A problem-centered, success-neutral common object Recommended, but requires explicit social, institutional, market, or technical modifiers
Coordination arrangements Mechanisms and interfaces without presuming deliberate design Recommended object label; adaptation must be specified rather than assumed

Collective agency: the sharpest conceptual rival

Collective agency is the strongest alternative when the research question concerns responsibility, authorized voice, rational group attitudes, or institutional personhood. It is much more discriminating than the coordination frame.

Criterion Coordination of interdependent activity Collective agency
Multiple contributions Required Required
Outcome-relevant coupling Required Normally present
Shared intention Not required Some group-purpose structure normally required
Stable decision procedure Not required Required
Authorized group voice Not required Required
Capacity for commitment Not required Required
Markets and crowds Included when coupled Generally excluded
Clocks and protocols Included functionally Excluded except by analogy
Failed coordination Directly diagnosable Diagnosable only within agent-capable groups
Legitimacy Additional variable Authorization may be represented, but moral legitimacy is not guaranteed

Collective agency is therefore a strong subset, not a viable umbrella.

Organized action: the strongest explanatory rival

The more important rival is the tradition of organized action, especially the Crozier–Friedberg lineage. Its formalizations apply beyond bounded organizations to “systems of organized action” and treat the management of power relationships as the core of interaction.27

That changes the explanatory question:

  • A coordination account asks how dependencies are managed.
  • An organized-action account asks how actors use control over dependencies and uncertainty to acquire, preserve, or contest power.

The first travels further across substrates. The second often explains more in human institutions. An algorithmic labor platform, for example, can be represented as a coordination arrangement that allocates work and records performance. But its evolution may be driven by managerial control, worker resistance, unequal exit options, and distributional conflict—variables the thin structural account deliberately brackets.

“Coordination of interdependent activity” is therefore the better comparative vocabulary. “Organized action” is the stronger rival whenever asymmetric power is causally central.

8. Implications and design possibilities

8.1 Use the frame to compare questions, not to transport answers

The framework can help researchers ask comparable questions across domains:

  • What is dependent on what?
  • What information is available where?
  • Which interfaces expose or conceal relevant state?
  • Who can allocate, sequence, veto, or revise?
  • What happens when a component fails, defects, or disagrees?
  • At what level are local and global choices composed?
  • Whose criterion defines success?
  • What evidence supports the mechanism claim?

It does not follow that an auction, hierarchy, consensus protocol, sanction system, or shared-memory design that works in one setting should be copied into another.

8.2 Separate operational performance from institutional evaluation

At minimum, social applications should report two evaluations:

  1. Operational: Did the arrangement manage its dependencies?
  2. Institutional: Was authority legitimate, were burdens and benefits distributed acceptably, and could affected actors contest or exit?

A system can score highly on the first and poorly on the second. Treating legitimacy as another coordination mechanism would erase that distinction.

8.3 Design experiments around boundary conditions

Several empirical programs follow plausibly from the evidence:

  • Predictive mechanism tests: Hold a dependency type constant while varying the proposed coordination mechanism across settings.
  • Failure-sampled comparisons: Study dissolved networks, failed commons, abandoned projects, demobilized movements, and ineffective emergency responses rather than only durable exemplars.
  • Dual coding for power: Describe the same arrangement both as dependency management and as a distribution of control, then test which variables explain adaptation and resistance.
  • Cross-substrate AI tests: Compare governance topologies under pre-registered noise estimates, hard single-agent baselines, agent failures, and multiple task regimes.
  • Semantic-equivalence tests: Determine whether constructs such as accountability, commitment, common ground, or collective choice retain comparable measurement properties in human–AI and machine-only systems.

These are research proposals, not established superior designs.

8.4 Prefer diagnostic repertoires to universal blueprints

Ostrom’s experience is instructive. Specific institutional rules did not transfer cleanly across sites, while more abstract lessons—clear boundaries, monitoring, conflict resolution, nested organization, and fit to local conditions—proved more durable within commons research.14 Even there, the principles are better treated as context-sensitive best practices than as a universal constitutional recipe.

The same restraint should apply elsewhere. The comparative vocabulary can identify relevant dimensions and missing assumptions. It should not promise that the same mechanism will produce the same result in a protocol, workplace, movement, commons, and AI collective.

9. Important uncertainties and limitations

1. Intentionality remains the central unresolved boundary

Whether coordinated action requires intentional joint orientation or merely causal coupling determines the extension of the entire framework. The structural choice made here is useful for cross-domain comparison, but the evidence does not establish it as the uniquely correct definition.

2. Predictive validity is unproven

Dependency and mechanism categories recur across several literatures, but recurrence is not prediction. No inspected study validates the proposed grammar across the eight problem families, and a prominent fast-response study rejects prespecified mechanism typologies as too formal for its setting.19

3. The core may still be too broad

Outcome-relevant interdependence can encompass competition, warfare, security conflict, epidemiological interaction, and ecological coupling. The boundary between coordination and strategic interaction remains unsettled.

4. System boundaries are observer-dependent

Markets, movements, communities, commons, and open-source ecosystems have porous boundaries and indirect effects. The framework does not provide a generally accepted rule for deciding which actors, activities, and dependencies constitute the focal arrangement.

5. Power is not just an optional module

The thin coordination definition does not deny power; it brackets it. But in coerced labor, organizational politics, platform control, and many public institutions, power may be the principal explanatory variable. A cross-domain vocabulary that omits it cannot be treated as a complete social theory.

6. Evidence does not transfer with terminology

Formal proofs, simulations, benchmarks, laboratory games, comparative case studies, and ethnographies establish different kinds of claims. Similar labels cannot make their effect estimates commensurable.

7. The comparative matrices are provisional

No direct study evaluates all eight problem classes through one operational framework. The transfer and functional matrices in this report synthesize heterogeneous evidence and should be tested rather than treated as settled taxonomies.

8. Failure evidence is sparse

The open-source mortality study is the clearest failure-sampled evidence, but it covers one domain and one sample. Failed commons, dissolved networks, demobilized movements, and ineffective emergency arrangements remain underrepresented.

9. AI evidence is immature and task-dependent

Recent studies show that organizational structure can affect multi-agent performance, but estimated gains vary sharply across tasks, and evaluation itself may depend on LLM judges. Longitudinal evidence about durable identity, responsibility, institutional learning, and high-stakes governance remains limited.

10. The corpus is uneven

Several influential works were available only as abstracts, snippets, or partial texts, including parts of the mobilization, organizational-learning, crowd, and organization-design literature. The source base is predominantly Anglophone, which limits any claim to universal institutional coverage.

Conclusion

“Systems for coordinated action” should not be established as a unified field. The phrase suggests a class of purpose-built systems producing successfully coordinated action, while the defensible extension includes emergent order, coercion, failure, markets, protocols, and processes without collective intention. Its component literatures study different dependent variables and support different kinds of inference.

The smallest coherent object is narrower:

Coordination arrangements managing outcome-relevant dependencies among distinguishable activities.

This object supports a useful comparative program. It provides a common exclusion rule, a vocabulary for dependencies and coupling mechanisms, and a disciplined way to compare assumptions across social and technical settings. It does not supply a universal theory of cooperation, governance, organization, agency, or adaptation, and it has not been validated as a predictive design grammar.

The best use of the frame is therefore translation, not subsumption. It can show where a market, trauma center, commons, protocol, movement, and AI collective confront structurally comparable dependencies. It should also show where the comparison ends—where incentives, intentions, legitimacy, power, interpretation, and evidence make the cases different.

That makes it neither a domain nor an empty umbrella. It is a vehicle for comparison whose usefulness depends on refusing to mistake commensurable description for common explanation.

Footnotes

  1. Kees van Kersbergen and Frans van Waarden, “‘Governance’ as a Bridge Between Disciplines: Cross-Disciplinary Inspiration Regarding Shifts in Governance and Problems of Governability, Accountability and Legitimacy,” European Journal of Political Research 43, no. 2 (2004): 143–171, https://doi.org/10.1111/j.1475-6765.2004.00149.x. Full text was available. ↩

  2. Thomas W. Malone and Kevin Crowston, “The Interdisciplinary Study of Coordination,” ACM Computing Surveys 26, no. 1 (1994): 87–119, https://doi.org/10.1145/174666.174668. Full working-paper text was available. ↩

  3. Francis Heylighen, “Stigmergy as a Universal Coordination Mechanism: Components, Varieties and Applications,” preprint of a chapter in Human Stigmergy (Springer), circa 2015–2016, https://pespmc1.vub.ac.be/Papers/Stigmergy-varieties.pdf. Full text was available. The paper’s own universality claim is theoretical rather than a cross-domain empirical test. ↩ ↩2

  4. “Collective Intentionality,” Stanford Encyclopedia of Philosophy, https://plato.stanford.edu/entries/collective-intentionality/; Abraham Sesshu Roth, “Shared Agency,” Stanford Encyclopedia of Philosophy, substantive revision May 1, 2017, https://plato.stanford.edu/entries/shared-agency/. Full entries were available. ↩

  5. Steven H. Strogatz, Daniel M. Abrams, Allan McRobie, Bruno Eckhardt, and Edward Ott, “Theoretical Mechanics: Crowd Synchrony on the Millennium Bridge,” Nature 438 (2005): 43–44, https://doi.org/10.1038/438043a. This source was available only at abstract and secondary-summary level, so the case is illustrative rather than decisive evidence for the broader boundary. ↩ ↩2

  6. Xavier Castañer and Nuno Oliveira, “Collaboration, Coordination, and Cooperation Among Organizations: Establishing the Distinctive Meanings of These Terms Through a Systematic Literature Review,” Journal of Management 46, no. 6 (2020): 965–1001, https://doi.org/10.1177/0149206320901565. The published article was inaccessible; the 11 percent result and definitions were verified through the authors’ own SAGE-published account of their findings. ↩

  7. Christian List and Philip Pettit, Group Agency: The Possibility, Design, and Status of Corporate Agents (Oxford University Press, 2011), accessed here through reviews and related listings; Roth, “Shared Agency”; Albert Bandura, “Exercise of Human Agency Through Collective Efficacy,” Current Directions in Psychological Science 9, no. 3 (2000): 75–78, https://doi.org/10.1111/1467-8721.00064, accessed at abstract level. ↩

  8. Philip Pettit, “Five Elements of Group Agency,” Inquiry (2023), https://doi.org/10.1080/0020174X.2023.2208391. Full text was available. ↩ ↩2

  9. Friedrich A. Hayek, “The Use of Knowledge in Society,” American Economic Review 35, no. 4 (1945): 519–530. Full text was available via Econlib. ↩ ↩2

  10. Katherine C. Kellogg, Melissa A. Valentine, and Angèle Christin, “Algorithms at Work: The New Contested Terrain of Control,” Academy of Management Annals 14, no. 1 (2020): 366–410, https://doi.org/10.5465/annals.2018.0174. Full text was available. ↩

  11. Michael J. Fischer, Nancy A. Lynch, and Michael S. Paterson, “Impossibility of Distributed Consensus with One Faulty Process,” Journal of the ACM 32, no. 2 (1985): 374–382, https://doi.org/10.1145/3149.214121. Full text was available. ↩

  12. “Arrow’s Theorem,” Stanford Encyclopedia of Philosophy, 2025 substantive revision, https://plato.stanford.edu/entries/arrows-theorem/. Full text was available. ↩

  13. Leslie Lamport, “Time, Clocks, and the Ordering of Events in a Distributed System,” Communications of the ACM 21, no. 7 (1978): 558–565, https://doi.org/10.1145/359545.359563. Full text was available. ↩

  14. Elinor Ostrom, “Beyond Markets and States: Polycentric Governance of Complex Economic Systems,” American Economic Review 100, no. 3 (2010): 641–672, https://doi.org/10.1257/aer.100.3.641. Full text was available. Ostrom reports the updated design-principle review by Michael Cox, Gwen Arnold, and Sergio Villamayor-Tomás, covering more than 100 studies; that review was not independently inspected. ↩ ↩2 ↩3

  15. Keith G. Provan and Patrick Kenis, “Modes of Network Governance: Structure, Management, and Effectiveness,” Journal of Public Administration Research and Theory 18, no. 2 (2008): 229–252, https://doi.org/10.1093/jopart/mum015. Full text was available. ↩

  16. Federal Emergency Management Agency, National Incident Management System, 3rd ed. (2017); Gregory A. Bigley and Karlene H. Roberts, “The Incident Command System: High-Reliability Organizing for Complex and Volatile Task Environments,” Academy of Management Journal 44, no. 6 (2001): 1281–1299, https://doi.org/10.5465/3069401. These materials were available only partially; the emergency-response classification should consequently be treated as provisional. ↩

  17. Yochai Benkler, “Coase’s Penguin, or Linux and the Nature of the Firm,” Yale Law Journal 112, no. 3 (2002): 369–446, https://arxiv.org/abs/cs/0109077. Abstract and partial full text were available. ↩

  18. Lyra J. Colfer and Carliss Y. Baldwin, “The Mirroring Hypothesis: Theory, Evidence, and Exceptions,” Industrial and Corporate Change 25, no. 5 (2016): 709–738, https://doi.org/10.1093/icc/dtw027. Full text was available. ↩

  19. Samer Faraj and Yan Xiao, “Coordination in Fast-Response Organizations,” Management Science 52, no. 8 (2006): 1155–1169, https://doi.org/10.1287/mnsc.1060.0526. Full text was available. ↩ ↩2

  20. Kevin Crowston, “A Taxonomy of Organizational Dependencies and Coordination Mechanisms,” MIT Center for Coordination Science Working Paper 174 (1994), https://crowston.syr.edu/sites/crowston.syr.edu/files/ATaxonomyOfOrganizationalDependencies.pdf. Full text was available. ↩

  21. Jailton Coelho and Marco Tulio Valente, “Why Modern Open Source Projects Fail,” in Proceedings of ESEC/FSE 2017, https://doi.org/10.1145/3106237.3106246; https://arxiv.org/abs/1707.02327. Full text was available. The study supports a conclusion about this sample of deprecated projects, not open-source mortality generally. ↩

  22. Chao Fei, Hongcheng Guo, and Yanghua Xiao, “When Agents Evolve, Institutions Follow,” arXiv:2604.27691 (2026). Full text was available. The paper treats historical institutions as design templates rather than literal equivalents of multi-agent systems. ↩

  23. Alibek T. Kaliyev and Artem Maryanskyy, “How Much Coordination Gain Is Real? A Paired Noise-Floor Protocol for Multi-Agent LLM Benchmarks,” arXiv:2606.20695 (2026). Full text was available. Its literature comparison is explicitly illustrative, heterogeneous, and not a universal statistical threshold. ↩

  24. Mert Cemri, Melissa Z. Pan, Shuyi Yang, et al., “Why Do Multi-Agent LLM Systems Fail?”, arXiv:2503.13657, NeurIPS 2025 Datasets and Benchmarks Track. Full text was available. ↩

  25. Shaida Kargarnovin, Christopher Ivan Hernandez, Dirk Reiners, et al., “From Testbeds to High-Stakes Work: A Review of Human-AI Teaming Domains and Teaming Factors,” Frontiers in Robotics and AI (2026), https://doi.org/10.3389/frobt.2026.1733942; Sophie Berretta, Alina Tausch, Greta Ontrup, et al., “Defining Human-AI Teaming the Human-Centered Way: A Scoping Review and Network Analysis,” Frontiers in Artificial Intelligence 6 (2023): 1250725, https://doi.org/10.3389/frai.2023.1250725. Full texts were available. The 2026 source is a review with a bibliometric component, not a pooled-effect meta-analysis. ↩

  26. James G. March, “Exploration and Exploitation in Organizational Learning,” Organization Science 2, no. 1 (1991): 71–87, https://doi.org/10.1287/orsc.2.1.71. This source was available only at abstract level. ↩

  27. Christophe Sibertin-Blanc, Pascal Roggero, Françoise Adreit, et al., “SocLab: A Framework for the Modeling, Simulation and Analysis of Power in Social Organizations,” Journal of Artificial Societies and Social Simulation 16, no. 4 (2013): 8, https://doi.org/10.18564/jasss.2278. Full text was available; it formalizes the organized-action tradition associated with Michel Crozier and Erhard Friedberg. ↩