Tacit knowledge and employee-AI collaboration: dynamic capabilities of the organization

Purpose As artificial intelligence (AI) tools become embedded in everyday work, employees increasingly engage in adaptive, experiential collaboration with AI systems. While such collaboration often generates valuable employee-level tacit knowledge, organizations struggle to translate this learning into strategic capabilities. This study examines how organizations can harness the knowledge emerging from employee–AI collaboration to build knowledge-based dynamic capabilities (KBDCs). Design/methodology/approach We conducted an inductive, qualitative study based on 29 semi-structured interviews with frontline employees, middle managers and senior leaders across multiple industries. Findings We develop a process model showing how tacit knowledge generated through micro-level human–AI collaboration is externalized as articulated experiential insights, validated and simplified into shared heuristics, and ultimately codified and embedded in organizational routines. This bottom-up transformation may help firms to renew internal knowledge resources and enhance KBDCs. The process is contingent on enabling conditions such as managerial support, organizational culture and employees' perceived agency in interacting with AI. Practical implications Organizations seeking to build dynamic capabilities from AI use should create structures that facilitate articulation, validation and dissemination of employee-level AI insights. Leaders play a critical role in translating individual experimentation into collective learning. Firms can assess their position in the transformation process and invest in systems that support phase-to-phase transitions. Originality/value While prior research has emphasized acquiring codified or external knowledge for dynamic capabilities, this study shifts attention to the internal, tacit and evolving knowledge that arises from employee–AI collaboration. We advance theory by unpacking a bottom-up pathway of KBDC formation through which such knowledge is articulated, validated and embedded over time. In doing so, the study also identifies a recursive organizational learning mechanism suited for high-velocity, AI-enabled environments.