001 / ORGANIZATIONAL CONTROLVSMLITE

Inspired by Beer’s Viable System Model.
Modified for persistent agent organizations.

Organizing agentsis a systems problem.

01 / FRAGMENTATION

AGENT HARNESS LANDSCAPE

1,000+ harnesses.No shared language.

Similar responsibilities arrive under different names, boundaries, and diagrams. Vocabulary hides the relationships we need to compare.

One term can describe different responsibilities. Different terms can describe the same responsibility.

02 / NORMALIZATION PROBLEM

BEYOND VOCABULARY

Renamingis not normalization.

A common dictionary cannot align authority, feedback, or responsibility. The comparison must be based on organizational relations.

REQUIRED / SHARED COORDINATE SYSTEM WHAT MUST A VIABLE
AUTONOMOUS SYSTEM
BE ABLE TO DO? FUNCTION · AUTHORITY · RELATION · FEEDBACK

We need a model that stays stable when implementations and names change.

REFERENCESCross-harness terminology examples ↗

03 / RELATION MODEL ↖

S5 / POLICYHold identity

Balance the present with the future.

MANAGEMENT

AUTONOMOUS OPERATIONS

EXTERNAL ENVIRONMENT

BALANCE STATUS / CONSTRAINTS DIRECT SAMPLE

BEER / VIABLE SYSTEM MODEL

ONE ARCHITECTURE · SELECT A CHANNEL · BEER / VSM ↗

04 / LAYERS ↖

04ORGANIZATION

responsibility · autonomy · continuity

OPEN METHODOLOGY

03VSMLITE

recursive control · synthesis · evolution

RESEARCH LAYER

02RELIABILITY

durability · telemetry · evaluation

Temporal

OpenTelemetry

LangSmith

01RUNTIME

state · tools · handoffs · workflows

LangGraph AGAgents SDK

AutoGen

00INFRASTRUCTURE

compute · isolation · scheduling

Kubernetes

Docker

E2B

05 / WHEN USE ↖

Decision tree

FAST GATE1,000+ CONCURRENT AGENTS?YES →USE VSM AS THE SHARED SYSTEM MAP

01 / OWNERSHIP

Does responsibility persist?

NO → ORCHESTRATE TASKSYES → S1

EST. 1–10+

02 / DEPENDENCE

Must autonomous units coordinate?

NO → KEEP S1 LOCALYES → S2

EST. 10–50

03 / CONTROL

Are resources and evidence shared?

NO → COORDINATEYES → S3 + S3*

EST. 50–400

04 / CHANGE

Must the organization adapt and choose policy?

NO → OPERATEYES → S4 + S5

EST. 200–1,000

FULL FORMALIZATIONS1 · S2 · S3 · S3* · S4 · S5

At 1,000+, the model becomes an organizational passport: one legible contract for everyone designing the system.

06 / AUTONOMY ↖

Less human support.
More autonomy.

OSM / ASSISTED VIABILITYSUPPORT → TRANSFER → INDEPENDENCE

THE HUMAN / PARENT Holds the whole.

WORKCOORDINATEREGULATEVERIFYADAPTGOVERN

Functions still supplied by the parent

THE EMERGING ORGANIZATION Begins with intent.

WORKCOORDINATEREGULATEVERIFYADAPTGOVERN

Functions owned inside the new system

Est. concurrent agents—Parent-led / intent

EXPLORE THE TRANSFER 00 / INTENT

The parent carries the missing functions.

A new organization starts inside an existing viable system. The human provides the support it cannot yet provide for itself.

As functions transfer, dependence on the human decreases. At full viability, the organization can sustain itself within its autonomy boundaries. OSM draft §6–7 ↗

07 / VSMLITE ↖

A small implementation
for testing the methodology.

VSMLITEEXPERIMENTAL PROTOTYPE / SEPTEMBER 2026

MEASUREcompletioncoordinationconflictsinterventionsrecoverylegibility

EXPLORES

recursive autonomy · stable roles · escalation · independent inspection · organizational synthesis

DOES NOT YET DEMONSTRATE

1,000-agent production deployment · organizational autoscaling · universal superiority over centralized orchestration

Back to products View VSMLite ↗Research →

REFERENCESStafford Beer — Brain of the Firm ↗Metaphorum — VSM ↗OSM draft ↗

08 / BONUS

Architecture
playground.

CONCURRENT AGENT SESSIONS1,000 / 1,000 RUNNING

S5GOVERN5

S4ADAPT25

S3REGULATE40

S2COORDINATE120

S1OPERATE795

S3* / VERIFY 15

INDEPENDENT S1795

CONTROL SESSIONS205

ACTIVE SYSTEMS6 / 6