While AI-powered agricultural advisory tools promise to address Nigeria's extension service gaps, adoption remains uneven and poorly understood. Quantitative studies document adoption correlates but cannot explain how farmers experience these technologies or why social structures produce differential access. This qualitative study explored the lived experiences of 42 smallholder farmers in Kaduna State, Nigeria, with an AI advisory tool, using the sample phenomenon of interest design evaluation research type (SPIDER) framework and semi-structured interviews and focus groups. Analysis through reflexive thematic analysis, grounded in Feminist HCI and Design Justice frameworks, revealed three key findings. First, farmers approached the technology with initial scepticism, evolving toward pragmatic integration where AI recommendations were layered onto indigenous knowledge rather than replacing tradition. Second, barriers operated across practical (phone charging, network coverage), social (gender norms, peer influence), and cultural (tradition, community decision-making) domains simultaneously. Third, gender relations, land tenure insecurity, and community hierarchies intersected to produce patterned inequalities, with women experiencing compounded disadvantage through limited phone access, time poverty, and restricted decision-making authority. The study is limited to one state and a specific AI tool, precluding statistical generalization. Findings imply that adoption is not a technical decision but a social process shaped by trust, social structures, and cultural meanings. We conclude that AI interventions must prioritize relational trust, structural inclusion, and cultural alignment. We recommend that policymakers invest in ongoing training with trusted extension workers, address gender inequality through design-by-inclusion approaches, strengthen land tenure security, and design tools for practical realities, including offline functionality and local languages.