Recent research has established that generative artificial intelligence (GenAI) is not a unitary workplace resource: exploratory and exploitative use orientations are associated with different forms of employee creativity, and their value depends on employees' capacity to integrate generated knowledge. This article therefore does not present the basic dual path as a new discovery. Instead, it develops a theoretical integration and dynamic extension that asks why the established paths operate differently, when employees should switch between them, and how short-run augmentation can evolve into either capability accumulation or capability substitution. We distinguish expansive and consolidative modes of knowledge integration, theorize phase-congruent switching rather than static ``high-high'' ambidexterity, and position metacognition as a control mechanism that governs both immediate knowledge conversion and longer-run cognitive development. We further specify cultural tightness and psychological safety as directional boundary conditions and distinguish individual creative gains from collective idea-space diversity. The framework converts emerging empirical regularities into falsifiable propositions about task phase, context, and time, providing a research agenda for sustainable human--AI creativity.