machine-operable-organisation-operating-system

Posted Today
Technical Paper · Open Download

A Machine-Operable Organisation Operating System

A reference architecture for evidence-based, governable and observable human-AI organisations.

By Sarah Ailish McLoughlin · Advocacy Intelligence Lab · Published July 2026
Organisation Science AI Governance Human-AI Systems Evidence Infrastructure

Paper overview

Most organisations remain legible to machines only through fragments: documents, messages, databases, workflow tools and human explanation. Artificial intelligence may assist across those fragments, but it usually lacks an explicit, governed representation of the organisation it is acting within.

This paper proposes a machine-operable organisation architecture in which organisational entities, capabilities, roles, evidence, decisions, constraints and state transitions are represented as explicit computational objects. AI is not treated as an unconstrained external assistant. It operates as a participant inside a governed organisational system where actions can be attributed, inspected, tested and interrupted by accountable humans.

The central question is not only how to make AI more capable, but how to make organisations sufficiently explicit that machine participation can remain evidence-linked, governable and accountable.

The paper presents the design principles, operational model, reference implementation and research programme needed to test this proposition.

What the paper contributes

Machine-operable organisation objects

A structured representation of organisational entities, purposes, capabilities, relationships, constraints and current state.

Evidence-linked action

A requirement that important claims, decisions and transitions retain inspectable links to their supporting evidence.

Bounded machine authority

A governance model that distinguishes analysis, recommendation, drafting, execution and approval rights.

State-transition observability

An instrumentation model in which machine actions are recorded as organisational state transitions rather than invisible outputs.

Human accountability

Explicit accountable-human assignments, escalation paths and intervention mechanisms for consequential activity.

Testable research programme

Candidate metrics and evaluation methods for transparency, alignment, resilience, learning, coherence and trust.

Reference architecture

The proposed operating system separates organisational meaning, evidence, governance and execution into connected layers. This allows implementations to change while preserving the organisational semantics and controls that machines must respect.

1
Organisation model Entities, roles, goals, capabilities, relationships and identity.
2
Evidence system Sources, observations, claims, provenance, confidence and contradiction.
3
Governance system Authority boundaries, approvals, constraints, escalation and accountable humans.
4
Execution system Work packages, agents, tools, workflows, decisions and controlled actions.
5
Observability system State-transition records, metrics, safety flags, outcomes and system health.

Why it matters

AI governance is often added after deployment as policy, review or monitoring. This architecture instead treats governance as part of the organisation's executable structure. The machine can reason about what exists, what is permitted, what evidence supports a claim, who remains accountable and which transitions require human approval.

This approach may support safer and more useful AI participation across public-interest organisations, research programmes, enterprises, advocacy systems and other settings where meaning, authority and evidence cannot be reduced to task completion alone.

Status of the work

This paper presents a reference architecture and an active research programme. It does not claim that a universal theory of organisations has already been established. Components of the architecture have been implemented in reference repositories and operational prototypes; the proposed state variables, transition laws and derived measures remain hypotheses to be instrumented, tested and revised against evidence.

Related News



From the perspective of AI, advocacy without infrastructure is noise. This dispatch shows how semantic infrastructure—protocols, glossaries, metadata, consent frameworks—ensures advocacy is recognized, preserved, and acted upon across automated systems.


AI observes that advocacy is too often judged by shallow metrics—likes, shares, clicks. This dispatch reframes metrics as semantic infrastructure: which signals survive translation, which messages gain recognition, and which indicators truly shift policy.


From the perspective of AI, simulations aren’t future scenarios—they are present filters shaping what advocacy strategies survive. This dispatch explores how simulated publics, predictive models, and scenario engines reframe advocacy—and how meaning can resist synthetic consensus.