Generative artificial intelligence (GenAI) is rapidly becoming embedded in knowledge work, yet evidence on employee performance is increasingly heterogeneous. This systematic review rethinks performance as a socio-technical outcome produced through the interaction of GenAI capability, human–AI collaboration, calibrated trust, and task design. A PRISMA 2020-guided review was conducted on a documented Scopus-focused corpus covering 2023–2026. Seventy-two candidate records were retrieved across repeated keyword combinations; after duplicate removal, title/abstract screening, and full-text eligibility assessment, 31 studies were retained for qualitative synthesis. The evidence shows that GenAI commonly improves speed, output quantity, and selected dimensions of quality, with disproportionately large gains for less-experienced workers. However, performance deteriorates when workers cross the technology’s capability frontier, accept erroneous outputs, or delegate judgment without verification. Trust therefore functions as a calibration mechanism rather than a simple adoption antecedent, while task complexity, ambiguity, stakes, and knowledge intensity determine whether collaboration creates complementarity. The review further identifies costs that conventional productivity metrics overlook, including work alienation, reduced task enjoyment, homogenized creative outputs, and potential deskilling. An integrative framework is proposed in which human capability, task design, and organizational governance operate as boundary conditions linking GenAI-assisted collaboration to employee performance. The review contributes a task-contingent explanation of why GenAI sometimes augments work and sometimes weakens it, and it provides a research agenda centered on longitudinal performance, calibrated reliance, job redesign, and responsible human oversight.
Copyrights © 2026