From Model Design to Organizational Design: Complexity Redistribution and Trade-Offs in Generative AI
Citation. Hasan, S., Oettl, A., & Samila, S. (2025). From model design to organizational design: Complexity redistribution and trade-offs in generative AI. SSRN. https://doi.org/10.2139/ssrn.5287538
Summary. Hasan, Oettl, and Samila develop the Generality-Accuracy-Simplicity (GAS) framework for reasoning about generative AI in organizations. They argue that a simple, general-purpose user interface does not remove complexity: it transfers it to infrastructure, data and compliance systems, specialized personnel, and workflow design. The paper treats this redistributed complexity and the resulting accuracy ceiling as organizational-design problems rather than merely model-performance limitations.
Key claims
- The GAS framework holds that generality, accuracy, and simplicity cannot all be maximized without trade-offs; GenAI’s apparent combination of broad capability and simple use is sustained by abstraction layers that relocate complexity rather than eliminate it.
- Where the cost of error is high, organizations may need to trade generality or interface simplicity for specialized models, retrieval and other infrastructure, domain expertise, and human verification.
- The paper argues that competitive advantage depends less on generic AI adoption than on designing workflows, abstraction layers, complementary expertise, and governance that can carry the hidden complexity of reliable use.
Evidence & limitations
- This is a conceptual working paper that synthesizes historical model-design arguments, technical literature, and illustrative organizational cases; it does not report a new causal study testing the GAS framework or its proposed organizational outcomes.
- The canonical DOI resolves to the 2025 SSRN record; the ingested Drive binary exactly matches the public arXiv v2 PDF (arXiv:2506.22440, updated 2026-05-15), so the source record uses the DOI for canonical provenance while retaining the hash of the ingested public version.
- The framework supplies a useful design hypothesis, but its claims about complexity redistribution, complementary capability, and competitive advantage require field evidence across organizations and task-risk conditions.