The production side of AI. The dominant question in organizational research asks what AI does to organizations. We ask the prior question: what did institutions do to AI before it arrived? AI systems are produced at positions within institutional worlds — shaped by organizational form, regulation, training data, feedback labor, evaluation regimes, market positioning, funding, and lineage — and we argue that they carry those positions into deployment. On that argument every AI arrives as a socialized entity, and choosing one is choosing the institutional world it carries. Our line: AI selection is institutional selection.
This program is one expression of a larger project. Intelligent Agency exists to theorize AI as one more intelligent entity in this universe, and to bridge what we know of human intelligence and human agency across to it where that knowledge applies. The production-side question below is what that bridge looks like when it is made empirical: entities are shaped by where they come from — so we go and measure whether these ones are, and how. The philosophical grounding is at The Journey.
The Throughline
Every paper in this program deepens a single argument: when organizations adopt AI, they are not adopting a tool. They are inviting in a socialized entity that carries the logics, priorities, and bounded rationalities of its originating social system.
This matters because organizations currently treat AI selection as a technical decision. It is, in fact, a social one — as consequential as hiring a person who was trained in a different institutional culture.
An AI system is not a thing apart. It is an intelligent entity that came from somewhere, and it carries where it came from. That is a claim about what AI is — and it is the claim these papers test.
Nonhuman Attentional Control: Organizational Import of External Institutional Logics
Academy of Management Annual Meeting 2026 · Organization and Management Theory (OMT) Division · Academy of Management Proceedings 2026(1) · doi.org/10.5465/amproc.2026.18749abstract
Do AI models carry institutional logics from their originating social systems? Given the same organizational scenarios, different AI models produce systematically divergent outputs, and each model repeats its own choices across runs — so the divergence is stable rather than random. Whether production origin, rather than capability that co-varies with it, is what causes the divergence is what the registered, capability-matched study tests.
- The five models diverged on 68% of the 25 scenarios, while each model repeated its own modal choice on 91.4% of its runs — the spread is between models, not within them
- 5 models × 25 organizational scenarios × 25 iterations = 3,125 observations
- Two models produced in the same country can diverge more than two produced in different countries: the divergence is not organized by geography
- AI attention is systematic, not random — the patterns are producer-linked, and what produces them is what the registered study tests
The Institutional Origins of AI: Evidence from Large Language Models
Develops the production-side framework: eight dimensions of production position (organizational type, regulatory context, training data, human feedback labor, evaluation regimes, market positioning, funding structure, distillation lineage), with two empirical studies showing institutional signatures reach into what models treat as evidence and explanation.
Study 1 is the decision study above (five models, 25 scenarios, 3,125 observations). Study 2 puts six open-ended knowledge tasks to nine models (162 outputs): the producer signature is faint in surface style and rises sharply in argument structure and knowledge organization — the model's share of variance grows from η² = 0.10 to 0.27 — and a cross-firm AI judge panel plus a deterministic parser corroborate the key findings.
What the two studies do not settle: with five producers, the production dimensions co-vary with one another and with capability. The paper does not claim that production origin rather than co-varying capability causes these differences; that is what the registered, capability-matched design is for.
What AI Agents Carry into the Loop: Network Positions of Production and the Reconfiguration of Epistemic Cultures
A theoretical companion accepted at the Society for Social Studies of Science (Toronto, October 2026), examining how network positions of production reconfigure the epistemic cultures that adopt AI systems.
The research continues
This work has evolved into a pre-registered study now underway.
Theoretical Foundations
This work draws on Actor-Network Theory's symmetry principle (what matters is what actants do, not their essential nature), the Attention-Based View of organizations (Ocasio), institutional theory's concept of embedded logics (Thornton, Ocasio and Lounsbury), and the economic sociology of trust and embeddedness (Granovetter).
The synthesis is novel: AI entities are formed within social systems as human actors are — but through artifacts (training data, configuration, fine-tuning) rather than lived experience, and whether the two are equivalent in effect is what this program tests.
The research program builds on a broader body of work spanning organizational learning, knowledge networks, sociomateriality, and the study of human-AI collaboration in organizational settings.