Structural AI Implementation Science (SAIIS): A Theoretical Framework for the Generative Era

Conventional implementation science, anchored in CFIR, TAM, UTAUT, and Rogerian diffusion theory, was constructed around a tacit ontological assumption: that the technology under study is a bounded, version-stable artifact whose behavior is fixed at the moment of adoption. Generative AI violates this assumption structurally rather than incrementally. The intervention is agentic, its capability surface drifts on a weekly cadence, and its output is co-produced with the user rather than delivered to the user. The result is a persistent and growing gap between adoption metrics and realized firm-level outcomes, what we term the implementation gap of the generative era. This paper proposes Structural AI Implementation Science (SAIIS) as a successor framework. SAIIS rejects the cognitive-individualist and intervention-static assumptions of legacy models and reorients the unit of analysis from user acceptance to organizational structural fit. We formalize three constructs that together specify the framework's empirical content. First, Floor-Lift Asymmetry (FLA) describes the structural dynamic by which generative systems elevate the baseline of cognitive deliverables without raising the ceiling of expert judgment, thereby invalidating the proxy signals (credentials, deliverable polish, response latency) on which prior labor markets and quality-assurance regimes depended. Second, the Thin-Shell Company (TSC) is defined as an emergent organizational archetype in which AI infrastructure carries at least 70 percent of core operational flows and the human layer is minimized rather than scaled. Third, the Dual-Track Talent Architecture (DTA) specifies the human-capital configuration, comprising Generalist AI Talent (GAT) and Composite Talent (CT), required to overcome the inertial friction that legacy organizations exhibit during structural transition. We further argue that the standard ROI calculus collapses under generative conditions: cost-reduction remains calculable-closed (computation yields decision), whereas efficiency-gain becomes calculable-open (computation requires a second, non-formalizable judgment). This asymmetry is not a measurement problem but a structural feature of the technology, and it has direct implications for how firms should sequence their transformation efforts. SAIIS thus reframes AI implementation as a problem of organizational form rather than user behavior, with consequences for management theory, labor economics, and the design of evaluation regimes.