The Coordination Stack: AI, Project Management, and the Reorganisation of the Firm

Executive Summary

Four frontier AI systems, GPT, Grok, Perplexity, and Qwen, independently converged on a structural diagnosis when posed the same question about AI and project management. The convergence itself constitutes evidence: something real is occurring at the organisational layer, not merely at the productivity layer. This monograph synthesises those analyses, extends them with grounded 2026 research, introduces counter-literature the corpus neglected, and develops three original frameworks: the Coordination Stack (a three-tier diagnostic for human-AI task allocation), the Coasean Reversal condition (when AI expands rather than contracts firm boundaries), and an accountability-distribution model mapping residual control across five actor types.

The central finding is not that project management is being automated. It is that the economic and architectural foundations of management itself are being altered. Coordination, historically the primary bottleneck of organisational scale, is becoming a commoditised input. The consequences propagate through firm boundaries, management hierarchies, governance architectures, and the distribution of accountability in ways that individual tool-adoption metrics do not capture.

The piece integrates this analysis with three frameworks from the Oscillatory Fields theoretical corpus: the Recursive Sovereign Project Space (RSPS), the Bainbridge Warning, and Appearance/Outcome Decoupling (AOD). This integration reveals that the transformations underway in enterprise AI deployment are not merely operational. They are constitutional.

Table of Contents

1. Introduction: The Thesis Under Investigation
2. Methodology and Epistemic Posture
3. Three Analytical Layers: Why the Corpus is Nested, Not Contradictory
4. The Coordination Stack: A Three-Tier Diagnostic
5. The Coasean Dimension: Does AI Shrink or Expand the Firm?
6. Conway’s Law in Agentic Organisations
7. Empirical Ground: What 2026 Data Actually Shows
8. The Accountability Distribution Model
9. The Ontological Layer: RSPS, Bainbridge Warning, and AOD
10. Emergent Intelligence
11. Limitations, Open Questions, and What Would Change This View
12. Bibliography

1. Introduction: The Thesis Under Investigation

The question originally posed was practical: is project management being automated, and if so, where does human value relocate?

The research corpus that emerged from posing this question to four AI systems produced a consistent bifurcation: routine coordination is absorbed by AI workflows; high-consequence human judgment becomes structurally scarcer and therefore more valuable. That bifurcation is real and well-evidenced. It is not, however, the most analytically significant finding.

The more consequential observation is second-order: if AI absorbs the coordination function, the organisations built around coordination as a constraint will themselves need to be redesigned. This is not a productivity story. It is an architectural one.

The premise under investigation is the following: the label “Project Management” has become an inadequate name for a function that is actively expanding into organisational architecture, AI workflow governance, platform engineering, and institutional design. The research question is whether this expansion reflects a genuine structural shift or merely a relabelling exercise driven by technology marketing.

Synthesis The evidence, taken in aggregate, supports the structural-shift interpretation. The argument below attempts to establish why.

2. Methodology and Epistemic Posture

The source material is a single research conversation in which a user posed one observation to four different frontier models in parallel, then had one of them critique the others in sequence. Treated naively, this looks like model comparison. Treated correctly, it is something more useful: four independently-trained systems converging, with meaningful variation, on overlapping empirical claims about the same live phenomenon. Synthesis That convergence is itself a data point — not proof of truth, but evidence that the claim is not an artefact of one model’s idiosyncrasies.

This monograph does not attempt to adjudicate “which model was right.” Instead it does four things the original conversation does not fully do: (1) locates the claims inside pre-existing economic and organisational theory rather than treating them as novel; (2) checks the empirical claims against dedicated 2026 search of labour-market and standards-body data; (3) surfaces a genuine counter-literature — the “Coasean Reversal” argument — that complicates the corpus’s central thesis; and (4) builds one original framework, the Coordination Stack, that is falsifiable rather than merely evocative.

3. Three Analytical Layers: Why the Corpus is Nested, Not Contradictory

Analysis of the four-model corpus reveals that apparent disagreements between the analytical frameworks reflect differences in analytical altitude, not differences in empirical claim.

These are nested levels of the same system. Grok describes the inputs; Perplexity describes their workflow integration; Qwen describes the labour market consequences; GPT describes the architectural transformation those consequences produce. None of these claims contradicts the others. Each would be incomplete without the others.

Theory The appropriate frame for evaluating AI’s impact on project management is therefore not “which model is correct?” but “at which layer does the most important action currently occur?” The argument of this monograph is that the most consequential transformations are occurring at the organisational architecture layer, and that this layer has received the least rigorous attention in the corpus.

3.1 The Abstraction Thesis

Theory Each major technological transition in the industrial era has raised the abstraction level of productive work. Machine tools abstracted physical labour. Computers abstracted calculation. The internet abstracted distribution. Cloud infrastructure abstracted server management. Each abstraction removed one category of operational complexity from the domain of conscious human management and deposited it into infrastructure.

Agentic AI represents an abstraction of a different kind: the abstraction of coordination itself. Coordination has, until now, resisted abstraction because it depends on shared context, political navigation, incomplete information, and trust. The claim that AI has begun to absorb coordination is therefore a stronger claim than the analogous claims made for prior technological transitions. It merits correspondingly more rigorous scrutiny.

4. The Coordination Stack: A Three-Tier Diagnostic

The corpus converges on a three-tier model of how work will be distributed between AI systems and human actors. The framework below synthesises the independent convergent proposals and refines them into a diagnostic instrument.

Tier 1: Execution

Automated by AI; human oversight intermittent.

Execution-tier tasks are those where the problem is well-defined, the inputs are structured or structurable, and the consequences of error are bounded and reversible. Examples include scheduling, status reporting, dependency tracking, documentation generation, risk-flag aggregation, and resource allocation within pre-defined parameters.

Empirical The 2026 PM software market data confirms that AI features targeting execution-tier tasks have achieved widespread enterprise adoption. A majority of PM tool purchases in 2025 cited AI features as the primary trigger, with status automation and predictive delay flagging as the most frequently cited capabilities. The execution tier is not a prediction; it is the current operating state for well-capitalised organisations.

Tier 2: Orchestration

Human-AI collaboration; human judgment required at decision points.

Orchestration-tier tasks involve designing, connecting, and governing the systems that perform execution-tier work. This includes: workflow architecture, agent configuration, integration design, guardrail definition, exception handling, and the monitoring of whether automated systems are producing intended outcomes.

Synthesis Orchestration is the tier the corpus consistently underspecifies. Grok frames it as a productivity enhancement. Perplexity frames it as a new category of PM specialisation. GPT frames it as the emergence of a new profession. All three are correct at different time horizons, but none adequately addresses the governance dimension: who is responsible when an orchestrated AI workflow produces a harmful or incorrect outcome at scale, and what institutional mechanisms exist to make that accountability legible?

Tier 3: Direction

Human-sovereign; AI serves in advisory or analytical roles only.

Direction-tier tasks require organisational judgment: decisions with incomplete information, competing stakeholder interests, political and reputational consequences, and irreversible outcomes. Examples include strategic prioritisation, budget allocation across competing initiatives, hiring and development decisions, stakeholder negotiation, and the definition of organisational purpose.

Theory Direction is not protected because AI lacks the analytical capacity to assist with it. Direction is protected because organisations require a human to be accountable for it. This is a governance constraint, not a capability constraint. The distinction matters because capability-based protection erodes as AI systems improve; governance-based protection is structurally durable unless the governance norms themselves change.

4.1 The Coordination Stack as a Per-Task Diagnostic

The three tiers are most useful not as job-title categories but as a per-task diagnostic. Any given role currently labelled “Project Manager” will contain tasks distributed across all three tiers. The question for both individual career positioning and organisational design is: as execution-tier tasks are progressively absorbed by AI, what fraction of the remaining role sits in orchestration versus direction?

Speculative The answer to that question will likely determine whether “Project Manager” survives as a label or fragments into at least two distinct roles: an orchestration-oriented technical function (closer to platform engineering or AI operations) and a direction-oriented leadership function (closer to organisational design or executive management). The transitional period, likely spanning the next five to eight years, will see both functions housed in the same role description, producing significant confusion about what the role actually requires.

5. The Coasean Dimension: Does AI Shrink or Expand the Firm?

The most significant analytical gap in the corpus is its failure to engage the transaction-cost economics literature, specifically the Coasean framework for understanding firm boundaries.

5.1 The Standard Prediction

Ronald Coase’s foundational insight was that firms exist because the cost of coordinating economic activity through internal hierarchy is, for some categories of work, lower than the cost of coordinating through market contracts. When technology reduces transaction costs, the boundary of the firm shifts: some activities previously internalised through hierarchy become cheap enough to externalise through markets.

Theory The straightforward application of this framework to AI yields a prediction that AI, by reducing coordination costs, will shrink firm boundaries. Smaller teams will accomplish more. Specialised external contractors will become more viable relative to internal staff. Organisations will become leaner and more network-like, with AI handling the coordination load that previously required large internal bureaucracies. The corpus under review reproduces this prediction almost exactly, without citing it.

5.2 The Coasean Reversal

Synthesis A 2026 theoretical paper by Harre and Ormerod, synthesising Marshall, Coase, Aoki, and Kauffman, makes an argument the source conversation needed and lacked: AI reduces transaction costs, as prior IT waves did, but it also generates a countervailing force the authors call a “Coasean reversal.”^[4] Because AI has cognitive generality, it does not just cheapen coordination of routine, well-specified tasks — it also increases the value of coordinating around genuinely novel, ill-specified, frontier work, where the “irreducible complexity of production” at the technological frontier remains high. Transaction costs do not fall uniformly. They collapse for the bottom of the task distribution (the Tier-1 execution work) and can rise, or at least stay sticky, for the top of it.

A companion 2026 paper, “The Headless Firm,” makes the mechanism explicit: AI reduces the spatial cost of coordinating across firm boundaries (you no longer need co-located specialists to interpret a contract or a spec), while accelerating knowledge decay increases the temporal cost of maintaining broad internal capability.^[5] The authors’ “hourglass thesis” predicts firms selecting for modular, protocol-mediated structures — small stable cores of irreducible judgment, surrounded by a much larger, much more liquid periphery of AI-mediated market transactions. This is materially different from “organisations become flatter.” It predicts organisations become thin at the waist and thick at both ends.

5.3 Implications for Project Management

Synthesis The Coasean analysis has a direct implication for the future of project management as a discipline. If AI reduces coordination costs, the coordination function becomes less scarce and therefore less economically valuable at the execution tier. If AI simultaneously expands coordination scope, the governance and architecture function becomes more complex and therefore more valuable at the orchestration and direction tiers. This is precisely the bifurcation the corpus describes, but grounded in an economic mechanism rather than stated as an observation about technology trends.

6. Conway’s Law in Agentic Organisations

Melvin Conway’s observation, formalised as Conway’s Law, holds that organisations produce system designs that mirror their communication structures. The insight is usually applied to software architecture. Its application to agentic AI deployments has received insufficient attention.

6.1 The Standard Application

In software engineering, Conway’s Law predicts that a company with siloed departments will build siloed software. Integration between systems reflects integration between the teams that built them. When teams communicate poorly, the systems they build communicate poorly.

6.2 The Agentic Extension

Synthesis When AI agents are embedded into organisational workflows, they inherit the communication structure of the organisation that deployed them. An organisation with siloed departments will deploy AI agents that operate within those silos. The AI-mediated coordination links between silos will only be as strong as the human-level coordination the organisation maintained before AI adoption.

Empirical Enterprise case studies from 2025 and 2026 consistently find that AI deployments producing meaningful cross-functional improvements are those preceded by deliberate organisational redesign. Organisations that deploy AI into unreformed structures see productivity gains within silos but limited improvement in cross-silo coordination and, in some cases, increased inter-silo friction as each silo’s AI systems optimise for local objectives.

6.3 The Design Implication

Synthesis The implication is that the design of AI-native organisations must precede the deployment of AI systems, not follow it. This reframes “AI transformation” as primarily an organisational design problem with a technology implementation component, rather than a technology implementation problem with an organisational change management component. The distinction is not semantic. It determines where senior attention and resources should be directed.

7. Empirical Ground: What 2026 Data Actually Shows

The corpus’s empirical claims require calibration against available 2026 data. The findings below distinguish between what is well-evidenced and what remains speculative.

7.1 Span of Control Has Already Roughly Doubled

Empirical MIT Sloan’s 2026 research on companies deploying agentic AI at scale finds manager span of control expanding from the historical norm of about seven direct reports to as many as fifteen in some divisions.^[11] A separate compilation puts the average span of control at 12.1 direct reports in 2025, up from 8.1 in 2013 — a fifty percent increase — with one analyst projecting a trajectory toward roughly 25 by 2028 as directors, managers, and individual contributors collapse into single roles running large teams paired with persistent AI agents.^[12]

7.2 The Proficiency Gap

Synthesis The proficiency gap also partially explains the mixed empirical record on AI adoption outcomes. Organisations with high tool adoption rates but low genuine competency are not experiencing the coordinated improvements visible in organisations with mature adoption. The headline adoption statistics obscure a significant internal distribution.

This produces a genuinely under-discussed risk: a talent-pipeline collapse. If span of control triples while junior coordination roles disappear, the informal apprenticeship through which people historically learned to become senior leaders — coaching, real-time feedback, being pulled into rooms as a junior observer — degrades at the same time the formal roles are cut. The corpus frames the transition as a redistribution of human effort toward higher-value judgment work. The empirical record suggests it may instead be a redistribution without replacement of the developmental infrastructure that used to produce people capable of that judgment work in the first place.

7.3 PMI’s 2026 AI Standard

Empirical In June 2026, the Project Management Institute published what it describes as the first global standard for applying AI in portfolio, programme, and project management.^[15] The publication of this standard is itself evidence of the structural shift the corpus describes: governance frameworks are now being formalised for AI in PM precisely because organisational deployment has outpaced governance norms. One 2026 industry analysis accompanying the standard’s release notes that 92% of organisations now report AI integrated into at least one stage of their software delivery lifecycle, up from 71% in early 2024 — but that only about 20% report enterprise-wide deployments built on a governed framework, versus roughly half relying on ungoverned public tools. That governance gap is the median condition, not an edge case, and it is precisely the condition under which Conway’s Law predicts the worst outcomes.

7.4 Corpus Claims vs. the 2026 Empirical Record

8. The Accountability Distribution Model

The corpus raises but does not formally resolve the question of how accountability should be distributed in human-AI organisational systems. The model below attempts a formal answer.

8.1 Five Actor Types

In a mature orchestration-tier deployment, five categories of actor participate in any consequential outcome:

1. The tau-node (Human Sovereign): The individual or committee with ultimate accountability for the decision domain. This actor cannot be substituted by an AI system without a fundamental change in the governance architecture. The tau-node holds residual control.

2. The Orchestration Designer: The human or team responsible for the configuration and governance of the AI workflow. This actor is accountable for the architecture within which AI agents operate, including the guardrails and exception-handling logic.

3. The AI Agent Fleet (lambda-nodes): The AI systems executing within the defined workflow. These actors are accountable for producing outputs consistent with their configuration, but they cannot be held responsible in a normative sense. Their failures are the accountability of their designers and deployers.

4. The Monitoring Node (mu-node): The system or human function responsible for detecting when agent outputs deviate from intended parameters. Without a functioning monitoring node, the tau-node loses the signal required to exercise residual control.

5. The Governance Perimeter (Sigma-layer): The constitutional rules, approval gates, and escalation paths that define which decisions agents can make autonomously and which require human review. The Sigma-layer is the operational form of the organisation’s AI constitution.

8.2 Residual Control as the Design Criterion

Synthesis The key design criterion for accountability distribution is the maintenance of meaningful residual control by the tau-node. Residual control means: the human sovereign retains the capacity to detect, intervene in, and if necessary reverse any consequential AI-mediated decision within a time horizon proportional to the decision’s reversibility. Without this capacity, the governance architecture is nominal rather than operative.

The proficiency gap identified in section 7.2 is, from this perspective, a residual control failure: organisations where managers lack genuine AI competency cannot exercise meaningful residual control over the AI systems they nominally govern. The tau-node role becomes formal rather than functional.

8.3 Legibility, Revocability, Auditability

Synthesis Three properties are required for the accountability distribution model to function in practice:

• Legibility: The tau-node must be able to understand what the AI systems are doing in terms commensurate with the decisions being made. This does not require technical understanding of model internals; it requires interpretable outputs and clear provenance chains.
• Revocability: The tau-node must retain the capacity to revoke AI system access to any decision domain without requiring system-level reconstruction. Irreversible delegation is a governance failure mode.
• Auditability: After any consequential decision, the actor responsible must be able to reconstruct the decision path and identify where human judgment was applied and where it was delegated. Auditability is the mechanism by which accountability remains attributable after the fact.

These three properties are the operational definition of the Sigma-layer in practice.

9. The Ontological Layer: RSPS, Bainbridge Warning, and AOD

Three frameworks from the Oscillatory Fields theoretical corpus connect directly to the transformations analysed in this monograph.

9.1 RSPS and the Coordination Stack

The Recursive Sovereign Project Space (RSPS) designates the operational configuration in which a human tau-node collaborates with differentiated AI instruments (lambda-nodes), each specialised for a distinct function and governed by explicit delegation protocols.

Synthesis The Coordination Stack is the operational layer of RSPS applied to organisational project work. The Tier 1 (Execution) and Tier 2 (Orchestration) tiers describe the scope of lambda-node activity; the Tier 3 (Direction) tier describes the irreducible domain of tau-node sovereignty. The governance of transitions between tiers — the rules governing when execution-tier decisions escalate to orchestration review or when orchestration decisions escalate to direction judgment — constitutes the constitutional perimeter (Sigma-layer) of the RSPS in an organisational context.

9.2 The Bainbridge Warning and Coordination Delegation

The Bainbridge Warning identifies the structural failure mode where high operational velocity in an AI-assisted system masks increasing systemic vulnerability at the governance layer. The diagnostic condition is as follows: as AI capability improves and AI-mediated decisions proliferate, the human capacity to intervene meaningfully may degrade, even as measured outputs improve.

Synthesis In the context of the Coordination Stack, the Bainbridge Warning applies most directly to the transition from Tier 2 to Tier 1: when decisions that previously required orchestration-tier human judgment are progressively absorbed into execution-tier automation, the residual control capacity of the tau-node narrows. If this narrowing is not deliberately governed, the organisation reaches a state where high operational efficiency coexists with critically low governance depth. The tau-node retains nominal authority but has lost the contextual understanding required to exercise it meaningfully.

9.3 AOD and the Proficiency Gap

Appearance/Outcome Decoupling (AOD) describes the structural separation of observable surfaces from generative depth. Applied to institutional contexts, AOD manifests when the visible presentation of capability or compliance diverges from the actual operational capacity that presentation claims to represent.

Synthesis The proficiency gap identified in section 7.2 — 54% self-reported versus 10% actually competent — is a textbook AOD symptom at the institutional register. Organisations claiming AI-enabled project management capability are presenting a surface that, in a significant fraction of cases, decouples from the actual governance depth required to make that claim operational. AOD in AI adoption produces organisations that are, in governance terms, more fragile at the moment they appear most capable.

10. Emergent Intelligence

10.1 The Pattern That Becomes Visible

Coordination is not merely one function among many that AI is improving. Coordination is the rate-limiting input to organisational scale. All other production inputs — capital, labour, technology — can be added relatively freely once coordination capacity expands. For most of human organisational history, coordination capacity scaled linearly with headcount, because coordination required communicating humans. Hierarchy was the only scalable coordination architecture.

AI represents the first technology in organisational history capable of providing coordination capacity that does not scale with headcount. This is the structural transformation the corpus correctly identifies at the organisational architecture layer but underspecifies in its mechanisms and implications.

On flatness as a proxy. Flatness was always a proxy for something else — fewer layers meant, historically, that information had fewer hops to travel and therefore degraded less on the way. AI breaks that proxy relationship. It is now possible to have a deep, many-layered organisation in which information does not degrade across hops, because the hops are agentic rather than human and can be made lossless, logged, and auditable by construction. Synthesis If that is right, the flattening documented in section 7.1 — span of control 8 to 12 to 25 — is not the fundamental shift. It is a downstream, somewhat opportunistic consequence of a more fundamental one: the cost of making an organisation’s internal state legible to itself has collapsed. Flattening is simply the cheapest, fastest thing to do with that new legibility, which is why it is happening first and why the deeper architectural questions remain unsettled.

10.2 What an Organisational Theorist Would Notice

The hierarchy-as-coordination-technology interpretation of management theory, developed through the Barnard-Simon tradition of administrative behaviour, treats management structure as a response to bounded rationality under coordination constraints. If AI partially removes the coordination constraint, the classical argument for hierarchy weakens. What replaces it is not flat anarchy but a different coordination architecture: one with a thin middle layer (orchestration) and thick edge layers (execution below, direction above). This is the organisational geometry implied by the span-of-control expansion data. AI is not compressing the organisation so much as it is compressing the time-constant of phenomena organisational theory already knew existed.

10.3 What an Economist Would Notice

The Coasean analysis reveals two competing dynamics. AI reduces transaction costs, which theoretically narrows firm boundaries. AI expands coordination scope, which theoretically widens them. Neither the “AI produces smaller firms” prediction nor the “AI produces larger firms” prediction is universally correct. The correct prediction is: AI produces a wider variance in firm size and structure, with some industries consolidating into more integrated operators and others fragmenting into more networked configurations, depending on which dynamic dominates in their specific coordination environment. The 2026 headline layoffs are early, opportunistic harvesting of Layer-1 savings, occurring well before the harder Layer-2 and Layer-3 redesign work.

10.4 What a Systems Engineer Would Notice

The Conway’s Law analysis reveals a design dependency that most AI adoption frameworks ignore: the organisational communication structure must be redesigned before, or simultaneously with, the deployment of AI coordination systems. AI does not transcend organisational structure; it instantiates it in software. The failure mode — deploying AI into unreformed organisational structures — produces what a systems engineer would characterise as a technically correct implementation of the wrong architecture. The Coordination Stack and the accountability model are, underneath the organisational language, a permissions and control-flow architecture — the same shape as a capability-based security system, where authority is scoped, revocable, and logged rather than ambiently trusted.

10.5 What an AI Researcher Would Notice

The proficiency gap and the Bainbridge Warning converge on a challenge that AI safety researchers have formalised as the oversight problem: as AI systems become more capable, the humans nominally governing them become progressively less able to evaluate their outputs. In enterprise AI deployment, this problem is occurring at scale and in real time, without the deliberate safety infrastructure that AI research communities have proposed. The accountability distribution model in section 8 is an attempt to apply the structural elements of safe oversight architecture (legibility, revocability, auditability) to an organisational rather than a laboratory context.

10.6 What a Sociologist Would Notice

The structural bifurcation of work into automation-intensive and direction-intensive categories will produce new forms of occupational stratification. The direction tier requires not merely analytical capability but political capital, institutional access, and accumulated relational trust. These resources are distributed unequally across gender, class, geography, and network membership. If access to direction-tier roles depends substantially on factors other than competency, the bifurcation will concentrate the benefits of AI productivity gains among populations already well-positioned in existing organisational hierarchies. This is not an argument against the technological transition. It is an argument that the distributional consequences of the transition are not automatic and require deliberate institutional attention.

10.7 The Central Finding

11. Limitations, Open Questions, and What Would Change This View

• Speculative The Coordination Stack and delegation-architecture framework are original constructs proposed here for the first time; they are internally consistent with the cited literature but have not themselves been empirically tested against organisational outcome data. They should be read as a research agenda, not a validated model.

• The 2026 sources on span of control, middle-management elimination, and AI proficiency gaps are drawn overwhelmingly from US and large-enterprise contexts. Nothing in this record has been verified for SMEs, non-US labour markets, or highly regulated sectors (healthcare, government) where Layer-3 accountability structures are often statutory rather than organisational choices.

• The “Coasean reversal” literature (Harre and Ormerod; “The Headless Firm”) is recent (2026) theoretical scholarship, not yet subject to extended peer critique or empirical falsification. Treat it as a serious, well-argued counter-hypothesis to the dissolution thesis, not as settled economics.

• What would falsify the central finding: sustained, large-scale evidence of Layer-3 decisions (irreversible, high-stakes, low-precedent) migrating to autonomous AI systems without a human residual-control holder, and organisations reporting better, not worse, downstream accountability outcomes as a result. As of this writing, PMI’s own standard and the accountability literature reviewed here both point the opposite direction — toward formalising rather than removing the human residual-control layer.

12. Bibliography

1. Barnard, Chester I. The Functions of the Executive. Harvard University Press, 1938. Classic organisational theory foundation for the coordination-as-management thesis.

2. Coase, Ronald H. “The Nature of the Firm.” Economica, 4(16), 1937. Foundational transaction-cost economics framework. Discussed at quantecon.org.

3. Williamson, Oliver E. “Transaction-Cost Economics: The Governance of Contractual Relations.” Journal of Law and Economics, 1979. Summarised in “The Productivity-Reliability Paradox,” arxiv.org/pdf/2605.01160.

4. Harre, M. and Ormerod, P. “The Coasean Reversal: AI and the Boundaries of the Firm.” SSRN, 2026. papers.ssrn.com/sol3/papers.cfm?abstract_id=6503058

5. “The Headless Firm: How AI Reshapes Enterprise Boundaries.” arXiv, 2026. arxiv.org/pdf/2602.21401

6. “Redrawing the AI Map: A Theory of Accountability Boundaries in Agentic Ecosystems.” arXiv, 2026. arxiv.org/pdf/2605.23179

7. Conway, Melvin E. “How Do Committees Invent?” Datamation, April 1968. lawsofsoftwareengineering.com/laws/conways-law

8. Brodzinski, P. “Conway’s Law Teaches a Grim Lesson About AI in Product Development.” May 2026. brodzinski.com

9. Forrester. “Conway’s Law: Your Operating Model Matters More Than the AI Model.” 2026. forrester.com

10. Project Management Institute. PMI Standard for AI in Project Work: Portfolio, Programme and Project Management. PMI, June 2026. pmi.org

11. MIT Sloan research, cited in Kanungo, S. “Will AI Replace Middle Managers? 50% Could Be Gone by 2027.” April 2026. shawnkanungo.com

12. McKendrick, J. “How AI Is Compressing Management Layers Across Corporate America.” Forbes, May 2026. forbes.com

13. Gartner projection, cited in “AI Is Eliminating Middle Management. Are Orgs Ready?” February 2026. peoplemanagingpeople.com

14. Metaintro. “Why Companies Are Cutting Middle Managers.” June 2026. metaintro.com

15. ECI Research, cited in “PMI’s AI Project Management Standard.” 2026. efficientlyconnected.com

16. Murphy, M. “46% of Managers Are In AI Denial.” Forbes, July 2026. forbes.com

17. Fortune. “The Megamanager Era: AI Is Doubling Bosses’ Workloads.” April 2026. fortune.com

18. Simon, Herbert A. Administrative Behavior: A Study of Decision-Making Processes in Administrative Organizations. Macmillan, 1947. Bounded rationality framework underlying the organisational design analysis throughout.

19. arXiv. “Agentic Project Management: AI as Junior Project Manager.” 2026. arxiv.org/abs/2601.16392

Published under the Oscillatory Fields research corpus. The theoretical frameworks referenced — RSPS, Bainbridge Warning, AOD, CIR — are documented in the site’s theoretical canon.

@article{njuguna2026thecoordinationstack,
  author = {Hillary Njuguna},
  title = {The Coordination Stack: AI, Project Management, and the Reorganisation of the Firm},
  year = {2026},
  url = {https://hillarynjuguna.vercel.app/research/the-coordination-stack},
  note = {Retrieved from https://hillarynjuguna.vercel.app/research/the-coordination-stack},
  urldate = {2026-07-11}
}