Can the integration of generative AI into public administration ease administrative burdens in streetâlevel bureaucracy? This article examines this question through a 6âmonth organizational ethnography conducted within a local authority in Shanghai. We find that while generative AI may alleviate certain traditional burdens, it can also paradoxically reinforce existing ones or create new forms. These dynamics, aligned with Moynihan, Herd and Harvey's (2015) conceptual framework, unfold across the interrelated dimensions of learning, compliance, and psychological costs. Critically, we identify a new type of burdenâwhat we term interpretive costsâwhich emerges in frontline administrators' everyday policy implementation and can be significantly reduced by AI integration. Our findings further suggest that, whether AI reduces, intensifies, or generates new burdens, it inevitably leads to policy alienation, characterized by an amplified sense of dehumanization, loss of control, and diminished meaning in their work. Through the lived experiences of SLBs navigating AIâassisted tasks, this article extends our understanding of administrative burdens in the age of generative AI.