Essay: The Architecture of Complexity

Core thesis
Herbert Simon argues that many complex systems are best understood as hierarchical assemblies of simpler subsystems. Complexity does not arise as a single undifferentiated tangle but as layers of structure in which higher-level units are composed of lower-level units. This hierarchical organization, he maintains, is not merely a convenient description but a fundamental feature that makes complexity tractable: systems are often "nearly decomposable," meaning their internal interactions are stronger than their interactions with other modules, allowing subsystems to function with a degree of autonomy.
Simon frames this insight around the idea that the architecture of complexity is shaped by the relative strengths and timescales of interactions. When components within a module interact rapidly and strongly, and interactions between modules are weaker or slower, the behavior of the whole can be approximated by understanding modules largely in isolation and then accounting for their weaker couplings. That approximation enables analysis, prediction, and design in domains where full, brute-force treatment of every interaction would be infeasible.

Mechanisms and concepts
Near-decomposability is the central mechanism Simon uses to explain why hierarchies form and persist. He distinguishes strict decomposability, where subsystems are entirely independent, from near-decomposability, where subsystems are only loosely coupled. The latter permits a two-step approach: first study the internal dynamics of subsystems assuming other parts are fixed, then study the slower dynamics of interactions among subsystems. Time-scale separation thus becomes a practical and conceptual lever for simplifying complex problems.
Associated with these ideas are the notions of modularity and coarse-graining. Modules serve as units of stability and reuse: they can be recombined and modified without collapsing the entire system. Coarse-graining, the practice of summarizing fine-scale details into effective higher-level descriptions, captures how higher-level behaviors emerge from lower-level activity while omitting irrelevant minutiae. Such abstraction is essential for both scientific explanation and engineering design.

Examples and implications
Simon illustrates his thesis through a range of examples, from watches and clocks to biological organisms and social organizations. A watch's gears form subsystems whose internal dynamics dominate until slow adjustments between groups of gears become significant. Biological evolution similarly produces hierarchical structures because incremental changes that stabilize at one level become building blocks for higher-level organization. The result is both robustness, since failures are often contained within modules, and evolvability, because modular parts can be recombined to produce new functions.
Practically, the argument reshapes how designers, managers, and scientists approach complex problems. Emphasizing modular design reduces cognitive and computational burdens, supports parallel development, and simplifies maintenance. In scientific modeling, recognizing near-decomposability guides model reduction and multiscale analysis, allowing researchers to focus effort where interactions are most consequential.

Influence and legacy
"The Architecture of Complexity" has had enduring influence across disciplines. Its framing of hierarchical, nearly-decomposable systems anticipates and underpins modern ideas about modularity in engineering, component-based software architecture, systems biology's multiscale models, and organizational theory's division of labor. It also seeded later formal work on complex adaptive systems, network theory, and evolutionary design by explaining why complexity can be built up incrementally rather than emerging fully formed.
Simon's essay remains a touchstone for thinking about how structure, interaction strength, and timescales jointly determine what features of a system matter for explanation and control. Its core insight, that hierarchical organization both arises from and enables manageable complexity, continues to guide efforts to analyze, design, and evolve complex systems.