Artificial intelligence can bring timely crop, weather, pest and market advice to millions of Indian farmers. Yet the same technology can also reproduce existing inequalities when access depends on a smartphone, a stable signal, a formal land record or a language that the system understands. This article explains how unequal connectivity, gendered control of devices, invisible tenancy, biased datasets and weak accountability may create a new agricultural divide. It argues for a human-plus-AI model built on low-bandwidth access, locallanguage testing, recognition of actual cultivators, transparent uncertainty and accessible grievance redressal. Smart farming will be genuinely smart only when its benefits reach farmers who are currently least visible to digital systems.