As artificial intelligence (AI) becomes increasingly embedded in journalistic practice, debates often frame it as either a disruptive force or a tool for efficiency. This study moves beyond this dichotomy by examining how journalists themselves interpret, negotiate, and position AI within journalism. Drawing on 19 in-depth interviews with Chinese journalists, and informed by boundary work and boundary object perspectives, the study analyses how AI is selectively integrated, justified, and governed across journalistic processes. The findings show that AI is primarily confined to routine and low-risk tasks, while being excluded from core editorial functions involving judgement, meaning-making, and accountability. The study argues that AI intensifies journalistic boundary work because it challenges journalism not only operationally, but also epistemically and infrastructurally. Journalists therefore engage in layered forms of boundary negotiation that operate across task, discursive, and structural dimensions. At the same time, AI functions as a boundary object mediating relationships among journalists, technology companies, and governance institutions under conditions of asymmetric platform dependency and institutional control.