Near-Decomposability
Near-decomposability is the architectural property that makes complex systems buildable by bounded minds. Simon introduced it in 'The Architecture of Complexity' (1962) using the parable of two watchmakers: Hora assembles watches in a hierarchy of stable sub-assemblies; Tempus assembles watches as single sequences of a thousand steps. When interruptions occur, Hora loses only the current sub-unit; Tempus loses all prior progress. The parable illustrates why nearly decomposable structures — systems in which interactions within subsystems are strong and interactions between subsystems are weak — are ubiquitous in biology, organizations, and software. Such structures allow each subsystem to be understood, assembled, and maintained by a bounded mind without requiring that mind to hold the entire system in view. AI weakens near-decomposability by enabling individuals to operate across subsystem boundaries — the backend engineer building frontend interfaces, the designer writing features, the product manager prototyping — producing both integration gains (fewer handoff losses) and cognitive costs (evaluation spread across unfamiliar domains with thinner pattern libraries).
In the AI Story
The principle is general. Biological organisms are nearly decomposable: organs interact strongly within their own systems and weakly across them. Social organizations are nearly decomposable: departments coordinate intensively within themselves and connect across boundaries through defined interfaces. Software systems are nearly decomposable: modules encapsulate functionality and communicate through APIs rather than shared global state. The ubiquity is not coincidental — near-decomposability is the architecture that bounded rationality requires, because systems that lack it cannot be understood or maintained by finite minds.
The Orange Pill documents the weakening of near-decomposability in the AI age without naming it as such. Segal describes engineers reaching across domains, designers writing code, and the emergence of vector pods that integrate functions previously distributed across teams. The diagnosis is accurate but incomplete: what Segal describes is a weakening of near-decomposability in one dimension (functional domain) and the emergence of near-decomposability in another (by cognitive operation — generation versus evaluation).
The dissolution is not pure loss. Handoff losses between modules — the translation costs that occur when information passes from designer to engineer to user — are reduced when a single mind operates across the boundary. Integration gains are real. But bounded rationality imposes a price: the mind that spans multiple domains cannot hold any of them as deeply as a mind that specialized in one. The ant-on-the-beach problem recurs: the sophistication of the output reflects the tool's pattern libraries across the expanded domain, while the evaluative depth at any particular point has necessarily thinned.
Origin
The 1962 paper appeared in the Proceedings of the American Philosophical Society and became one of Simon's most influential. It established near-decomposability as a cross-domain structural principle — applicable wherever bounded agents build or maintain complex systems. The paper's core claim, that complex systems generally evolve into nearly decomposable forms because nearly decomposable forms are easier to construct and more stable under perturbation, has been supported by subsequent research in biology, software engineering, and organizational theory.
Simon's collaborator Albert Ando later extended the mathematical foundation, showing that nearly decomposable systems exhibit two-stage dynamics: short-term behavior dominated by within-subsystem interactions, long-term behavior dominated by between-subsystem interactions. This separation allows bounded agents to analyze the system at one timescale at a time rather than having to integrate across timescales simultaneously.
Key Ideas
Strong within, weak between. The defining structural property of a nearly decomposable system is that interactions within subsystems dominate interactions between them.
Bounded minds require it. Systems that are not nearly decomposable cannot be understood, built, or maintained by finite cognitive agents.
Hora's architecture is resilient. The parable of the watchmakers illustrates why hierarchical sub-assembly is the only viable strategy in interruptible environments.
AI weakens the property. When a single mind can operate across subsystem boundaries, the near-decomposable structure begins to dissolve, producing both integration gains and evaluation losses.
New boundaries are emerging. The decomposition by functional domain is being replaced by decomposition by cognitive operation — generation versus judgment, implementation versus direction.
Appears in the Orange Pill Cycle
Further reading
- Simon, 'The Architecture of Complexity' (1962)
- Simon and Ando, 'Aggregation of Variables in Dynamic Systems' (1961)
- Simon, The Sciences of the Artificial (1969, 1996)
- Carliss Baldwin and Kim Clark, Design Rules: The Power of Modularity (2000)