While generative artificial intelligence (GenAI) has the potential to enhance knowledge work, the psychological impact of its use is not fully known. Based on Social cognitive theory and Self-determination theory, this study explores whether the use of GenAI erodes employees' professional self-efficacy and subsequently their meaningful work, and whether AI-literacy helps to protect against this negative trend. The three-wave time-lagged survey of 350 knowledge workers in Pakistan was analysed using composite-based regression (PROCESS Model 7) with bootstrapping method, which was cross-validated with PLS-SEM. Professional self-efficacy (β = −.277) and meaningful work (β = −.123) decreased with the dependence on GenAI. Professional self-efficacy was a significant predictor of meaningful work (β = .479) and accounted for half of the total effect (indirect effect = −.133). The indirect effect was reduced when AI literacy was low (−.199), but it became less negative and significant at high AI literacy (−.068; index of moderated mediation = .066). Results were consistent across the control variables, the PLS-SEM and the method-variance checks. The study moves the discussion about GenAI productivity to one about psychological cost, clarifies the role of self-efficacy in the link between reliance and meaningful work and introduces the concept of AI literacy as a trainable organisational shield.