The shape of organisations when software is no longer the constraint

Research Question

Inside an organisation that requires software to be built, integrated, and maintained (including Commercial Off-The-Shelf (COTS) systems, Software-as-a-Service (SaaS) platforms, and bespoke-built systems), how much of that organisation exists purely to manage the shaping, prioritising, management, or tracking of that work, and how does the organisation's structure change when the cost of producing software approaches zero?

Supporting questions:

• What fraction of a typical software-dependent organisation's headcount exists primarily to coordinate, manage, or govern software delivery rather than to deliver it directly?
• Which specific roles and functions exist as proxies or translators between business intent and software execution?
• What does economic theory (particularly transaction-cost economics) predict about firm shape when one major production cost collapses?
• What historical analogues exist, automation of manufacturing, adoption of cloud computing, introduction of Enterprise Resource Planning (ERP), and what did they do to organisational structure?
• What does "software cost going to zero" mean in practice today (Artificial Intelligence (AI)-assisted development, AI coding agents, natural-language-to-code pipelines)?
• Which coordination and management functions could themselves be automated, and which require irreducible human judgment?
• What happens to Agile rituals, planning cycles, estimation, and roadmapping when throughput is no longer the binding constraint?

Findings

Executive Summary

Key Findings

1. [inference] Medium to high confidence: In mid-to-large software-dependent organisations, explicit coordination and governance roles are best estimated at at least the high teens of headcount and often roughly one quarter to one third once team-level, train-level, architecture, release, and management layers are counted together.
2. [inference] Medium to high confidence: Product Owner, Scrum Master, Business Analyst, project or programme manager, delivery manager, release manager, change manager, and intermediary solution-architecture roles exist largely because business intent and software execution are separated by translation, validation, batching, and queueing costs.
3. [inference] Medium confidence: The prior repository item 2026-03-10-nature-of-the-firm-coase-organisations.md, which summarises Coase and Williamson transaction-cost theory, together with this item's historical analogues implies that materially cheaper coding should reduce the hierarchy devoted to allocating scarce engineering labor while preserving and often concentrating the hierarchy that owns standards, compliance, risk acceptance, and residual decision rights.
4. [fact] High confidence: Enterprise Resource Planning (ERP), cloud, and low-code and no-code adoption all reduced some local execution work but replaced it with new central structures for standards, enablement, platform governance, shared services, or Centers of Excellence rather than eliminating coordination altogether.
5. [fact] Medium to high confidence: Current AI evidence shows large gains on routine software tasks and measurable reductions in project-management friction, but it does not justify the stronger claim that complex enterprise software delivery is already near-zero cost from idea to production.
6. [inference] Medium confidence: The first coordination work to compress is artifact transformation, including backlog drafting, routine status reporting, repetitive release administration, and simple compliance evidence gathering, while judgment-intensive work remains human-heavy for longer.
7. [inference] Medium confidence: Agile rituals, estimation cycles, and roadmap ceremonies are likely to thin sharply wherever they exist mainly to ration scarce developers, but they will persist wherever they encode commitments, dependency management, risk review, or regulated change control.
8. [inference] Medium confidence: The best-supported future organisational equilibrium is smaller software-execution cells around stronger platforms and guardrails, with regulated firms tending toward a dual-core model of AI-heavy delivery plus concentrated governance rather than managerial collapse, although the prior repository item 2026-03-12-ai-force-multiplier-ambition-expansion.md implies that expanded demand could preserve more coordination capacity than a pure cost-reduction model would predict.

Assumptions

• [assumption] The current public AI productivity results are a reasonable directional proxy for the next several years of tooling progress, even though frontier models and enterprise workflows will change quickly. Justification: Multiple independent sources already agree on routine-task acceleration and reduced coordination friction.
• [assumption] Mid-to-large organisations using Scrum- or Scaled Agile Framework (SAFe)-like patterns are representative enough to support a bounded estimate of the coordination layer. Justification: These frameworks are common in the kinds of organisations targeted by the question, but they are not a literal census of all firms.

Analysis

• [fact] Current software organisations name coordination roles explicitly because business intent, technical implementation, and organisational accountability are split across different people and time horizons, which is visible in Scrum and Scaled Agile Framework (SAFe) role definitions and in economy-wide bureaucracy estimates.
• [fact] Current AI evidence shows clear compression in routine coding, documentation, and some project-management-adjacent work, which supports the repository's prior 2026-03-23-software-factory.md conclusion that the bottleneck is migrating away from code production.
• [fact] Historical analogues from Enterprise Resource Planning (ERP), cloud, and low-code and no-code adoption show that cheaper local execution does not produce a coordination-free organisation; it produces a different coordination layer centered on standards, platforms, guardrails, and risk ownership.
• [inference] The strongest competing explanation is demand expansion rather than headcount compression: the prior repository item 2026-03-12-ai-force-multiplier-ambition-expansion.md argues that cheaper software can unlock more initiatives, which means total coordination demand may fall more slowly than translation work does.
• [inference] The best-supported strategic implication is therefore to redesign around thinner translation layers and clearer decision rights while assuming that governance, residual accountability, and risk acceptance remain necessary even when production technology becomes much cheaper.

Risks, Gaps, and Uncertainties

• [fact] No clean public dataset decomposes software-dependent firms into builder versus coordinator headcount, so the percentage answer is a bounded estimate rather than a census fact.
• [fact] Several seeded sources were inaccessible or unusable in this environment: the Coase Wiley page returned 403, the Gartner low-code page returned 403, and the seeded Standish and Wired links returned 404.
• [fact] Some AI-productivity evidence comes from vendor-published sources such as Cognition and Cursor, which are useful directional signals but weaker than controlled experiments or independent working papers.
• [fact] Public evidence remains much stronger for routine coding and documentation tasks than for complex cross-system, regulated, or politically contested enterprise change.

Open Questions

• [fact] What is the best direct empirical method for measuring builder versus coordinator headcount in large software-dependent firms?
• [fact] At what point do AI agents reduce not just coding effort but also the demand for project-management and architecture-intermediary roles in regulated enterprises?
• [fact] Which governance model replaces the investment board or project front door when software throughput is no longer the binding constraint?

sources

• [x] The Nature of the Firm - Ronald Coase (1937) — - seeded primary source; checked, but the Wiley page returned status 403 from the runner, so it served as a source pointer rather than directly quotable evidence here
• [x] Markets and Hierarchies - Oliver Williamson (1975) — - seeded source pointer; the Harvard University Press page was accessible only as book metadata, so it was used for framing rather than direct quotation
• [x] The Second Machine Age - Brynjolfsson and McAfee (2014) — - seeded source pointer; the Amazon page was accessible but not directly useful for evidence extraction in this environment
• [x] McKinsey: Unleashing developer productivity with generative AI, via McKinsey August 2023 explainer — - accessible McKinsey material citing the seeded article's task-speed results
• [x] GitHub Copilot productivity research (Kalliamvakou, 2022) — - controlled experiment on task completion speed and developer experience
• [x] Stripe Developer Coefficient summary — - accessible summary quoting Stripe's 2018 findings on maintenance burden and developer productivity
• [x] CHAOS Report - Standish Group — - seeded link checked; returned status 404 and is not used as evidence
• [x] Devin AI - Cognition Labs — - vendor primary claim about autonomous software-engineering capability; used with low evidentiary weight
• [x] Cursor — - vendor site checked; useful as a signal of market positioning and adoption claims, not as independent productivity evidence
• [x] The Mythical Man-Month - Fred Brooks (1975) — - seeded source pointer for coordination-cost logic; the Amazon page was accessible only as book metadata
• [x] Wired: "Vibe coding" and the future of software — - seeded link checked; returned status 404 and is not used as evidence
• [x] Harvard Business Review: Excess Management Is Costing the U.S. $3 Trillion Per Year — - economy-wide bureaucracy baseline
• [x] London Business School: Bureaucracy, where to liberate $3 trillion — - accessible restatement of Hamel and Zanini's core management-overhead numbers
• [x] a16z: Why Software Is Eating the World — - framing source for software as a dominant production factor
• [x] Gartner: low-code and no-code market and organisational impact — - seeded link checked; returned status 403, so Microsoft and low-code governance sources below were used instead
• [x] Massachusetts Institute of Technology (MIT) Sloan: Leadership and AI insights for 2025 — - accessible MIT Sloan Management Review gateway page covering AI governance and organisational design implications
• [x] Stack Overflow Developer Survey 2024 — - empirical baseline on AI adoption and time spent searching for solutions
• [x] Scrum Guide — - primary framework description of Scrum roles and responsibilities
• [x] Atlassian: Agile Scrum roles and responsibilities — - accessible explanation of Scrum team composition and role structure
• [x] Scaled Agile Framework: Agile Release Train — - scaled-delivery role structure and headcount band
• [x] Harvard Digital, Data and Design (D^3): The AI Revolution in Software Development — - accessible summary of the 2024 working paper "Generative AI and the Nature of Work"
• [x] Google Cloud: Identifying and tracking toil using Site Reliability Engineering (SRE) principles — - automation of repetitive operations work
• [x] Google Cloud: Transitioning a typical engineering ops team into an SRE powerhouse — - operations automation changes team shape rather than eliminating operations
• [x] Amazon Web Services (AWS) Prescriptive Guidance: Building a Cloud Center of Excellence within your organization — - cloud-era governance and operating-model analogue
• [x] Microsoft Power Platform Center of Excellence Starter Kit overview — - low-code governance analogue
• [x] Quickbase: Low-Code CoE, Governance for IT Directors — - low-code governance analogue
• [x] Kissflow: How to Set Up a No-Code Center of Excellence — - low-code and no-code governance analogue
• [x] Strategic Approaches to ERP Implementation — - ERP centralization and process-standardization analogue