In the era of artificial intelligence, translating AI-enabled insights into circular business model innovation (CBMI) remains a major challenge for construction firms because implementation depends on workforce transformation rather than technology adoption alone. Drawing on the dynamic capabilities view and the 4I organizational learning framework, this study develops a knowledge-to-skill conversion model linking AI capability (AIC), green knowledge management (GKM), labor skill transformation (LST), government support, and CBMI. Survey data from 498 Chinese construction firms were analyzed using PLS-SEM, multi-group analysis, necessary condition analysis, and fsQCA. The results show that AIC is positively associated with GKM, LST, and CBMI. After LST is included, the average direct relationship between GKM and CBMI becomes non-significant, whereas the serial AIC–GKM–LST–CBMI pathway remains significant, indicating that green knowledge is translated into innovation mainly through workforce enactment. Government support strengthens the conversion of both GKM and LST into CBMI by improving the institutional conditions for knowledge application. Segment analyses show that this mechanism is stronger in industrial construction and infrastructure/heavy civil engineering, while different construction segments reach high CBMI through distinct configurations of AI, knowledge, workforce, and institutional capabilities. Our findings offer practical insights for industry practitioners, demonstrating that building CBMI involves more than simply adopting digital technologies; it also encompasses micro-level mechanisms for reshaping the workforce’s skill structure—specifically, reskilling, work redesign and work adaptability—to enable the transformation of these technologies into innovation.

Cultural sustainability in intangible cultural heritage (ICH) design requires contemporary visual innovation to retain cultural meaning and involve heritage holders. This study examines how generative AI capability relates to culturally grounded innovation using cross-sectional data from 542 matched designer–inheritor dyads in Guangyuan. Partial least squares structural equation modeling, multi-group analysis, necessary-condition analysis and fuzzy-set qualitative comparative analysis were applied. Generative AI-enabled visual design capability was positively associated with ICH visual design innovation, directly and indirectly through heritage knowledge integration, cultural visual translation and the theoretically ordered knowledge integration–visual translation pathway. Higher designer–inheritor collaboration quality was associated with stronger relationships between AI capability and both knowledge integration and visual translation. Point estimates followed the proposed heritage-context differences, but the group contrasts were not statistically supported. Necessary-condition and configurational analyses identified combinations of technological, cultural and relational capabilities, with limited configurational coverage and sensitivity to calibration. Integrating the dynamic capabilities and knowledge-boundary perspectives, the study distinguishes knowledge interpretation from visual enactment and locates collaboration quality within these capability relationships. The findings inform culturally sustainable use of GenAI by connecting visual exploration with verified heritage knowledge, continued inheritor participation and cultural-boundary governance. Design teams should document cultural sources and permissions, involve inheritors in interpretation and review, and assess cultural recognizability alongside novelty. These project-level associations identify capabilities relevant to cultural sustainability; long-term heritage transmission and safeguarding outcomes require further investigation.