Virtuous or Vicious Cycle? : Generative AI and the Reconfiguration of Dynamic Capabilities in Large-Scale Software Enterprises

Sammanfattning

Background: Generative AI (GenAI) tools have been rapidly adopted across softwareintensive large-scale enterprises (LSEs), yet productivity gains remain uneven and the organizational processes behind effective integration are poorly understood. Existing applications of the Dynamic Capabilities View (DCV) to AI adoption remain largely conceptual, leaving managerial actions, routines, and learning processes empirically underspecified. Purpose: This study explores the underlying mechanisms that drive the development and reconfiguration of dynamic capabilities as software-intensive LSEs adopt and integrate GenAI tools, moving beyond static description toward the generative structures that produce observable adoption patterns. Method: Grounded in critical realism, the study uses a qualitative, cross-organizational design combining abductive and retroductive reasoning. Ten semi-structured interviews with software developers and team leads across diverse LSEs were analyzed through hybrid thematic coding followed by retroductive mechanism inference. Conclusion: Four generative mechanisms were inferred: external legitimacy as an adoption trigger, the gap between formal governance and informal practice, AI literacy as a stratifying mechanism, and recursive capability erosion. Together they show that the DCV's iterative cycle can operate not only virtuously but also degeneratively, where short-term productivity gains erode the expertise needed for sustained capability development. The study empirically specifies the DCV in the GenAI software context and reframes "red tape" in LSEs.