Organizational Patterns & Formal Models

A mathematics of organizations mapped onto task graph primitives

abstract

Workgraph's primitives — tasks, dependency edges, roles, motivations, agents, a coordinator, evaluations, and an evolve loop — are not arbitrary design choices. They map precisely onto well-established concepts from organizational theory, cybernetics, workflow science, and distributed systems.

This paper develops a vocabulary and framework — a "mathematics of organizations" — that helps users think rigorously about how to structure work in workgraph. It connects the system's concrete primitives to the theoretical traditions that explain why they work.

read the full paper

HTML Read in browser PDF Download PDF

key findings

The task graph is a stigmergic medium

Agents coordinate indirectly by reading and writing task state, exactly as ants coordinate via pheromone trails. No agent-to-agent communication is needed — the graph is the communication channel.

The execute-evaluate-evolve loop is autopoietic

The system literally produces the components (agent definitions) that produce the system (task completions that trigger evaluations that trigger evolution). This is Maturana & Varela's self-producing network and Argyris & Schön's double-loop learning — at once.

The coordinator is a cybernetic regulator

Operating an OODA loop, subject to Ashby's Law of Requisite Variety: the number of distinct roles must match or exceed the variety of task types, or the system becomes under-regulated.

Evaluations solve the principal-agent problem

The human principal delegates to autonomous agents under information asymmetry. Evaluations are the monitoring mechanism; motivations are the bonding mechanism; evolution is the incentive-alignment mechanism.

Role design is an Inverse Conway Maneuver

Conway's Law predicts that system architecture mirrors org structure. In workgraph, deliberately designing roles shapes the task decomposition and therefore the output architecture.

The provenance log is organizational memory

wg trace records the full causal chain of every workflow — not just current state (stigmergy) but how it got there. This is Luhmann's structural memory: the system's capacity to selectively remember and forget its own history.

Replay transforms memory into learning

wg replay re-executes past workflows with different parameters, enabling double-loop learning and counterfactual reasoning. Successful patterns become organizational routines — reusable functions extracted from traces.

theoretical frameworks

Stigmergy

The task graph as coordination medium — indirect coordination through traces left in a shared environment. How after edges and task status changes create the pheromone trails that guide downstream agents.

Workflow Patterns

Van der Aalst's 43 control-flow patterns mapped to workgraph. What after edges and structural cycles can natively express, and what requires coordinator logic.

Fork-Join, MapReduce & Scatter-Gather

The three parallel decomposition patterns. The planner → N workers → synthesizer topology as workgraph's fundamental parallel structure.

Pipeline & Assembly Line

Sequential processing patterns, throughput vs. latency trade-offs, and the Theory of Constraints applied to task graph bottlenecks.

Autopoiesis

The self-producing agency. Maturana & Varela's biological concept applied to how the evolve loop creates the agent definitions that create the task completions that trigger the evolve loop.

Trace as Organizational Memory

Luhmann's structural memory realized in wg trace. The three layers of memory: stigmergic (graph state), provenance (operation log), and functional (extracted routines).

Cybernetics & Control Theory

The coordinator as a cybernetic regulator. Ashby's Law of Requisite Variety applied to role design. Single-loop vs. double-loop learning.

The Viable System Model

Beer's five systems mapped to workgraph. The S3-S4 balance between operational management and intelligence/adaptation.

Principal-Agent Theory & Conway's Law

Information asymmetry in autonomous agent delegation. The Inverse Conway Maneuver: shaping output architecture through deliberate role design.

Team Topologies & Organizational Primitives

Stream-aligned, enabling, platform, and complicated-subsystem teams mapped to workgraph roles. Division of labor, span of control, and coordination costs.

primitive ↔ theory mapping

Primitive	Theoretical Load
`Tasks`	The universal unit of work — mapped by all 10 frameworks
`after` edges	Workflow Patterns, Fork-Join, Stigmergy, Conway's Law, Coordination Costs
`Structural cycles`	Workflow Patterns (loops), Cybernetics (feedback), Autopoiesis, Agency Theory (repeated games)
`Roles`	Team Topologies, Conway's Law, VSM (S1), Requisite Variety, Division of Labor
`Motivations`	Agency Theory (bonding), VSM (S5 policy), Cybernetics (constraints)
`Coordinator`	Cybernetics (regulator), VSM (S3), OODA Loop, Agency Theory (principal's delegate)
`Evaluations`	Agency Theory (monitoring), Cybernetics (feedback), VSM (S3* audit)
`Evolve`	Autopoiesis (self-production), Cybernetics (double-loop), VSM (S4), Agency Theory (incentive alignment)
`Trace`	Org Learning (memory), Autopoiesis (structural memory), Stigmergy (persistent traces)
`Replay`	Org Learning (double-loop), Autopoiesis (reproduction with variation), Evolutionary Theory