From pilots to plots: A critical review of artificial intelligence for smallholder agriculture in South Asia

Artificial intelligence (AI) is routinely presented as transformative for the several hundred million people in South Asia whose food security depends on farms of less than two hectares, yet the claims made on its behalf have advanced faster than the evidence behind them. This review critically examines what is known about AI applications intended to reach smallholder farmers in the region, spanning advisory, diagnostic, precision, credit, and market uses. A corpus of 31 peer-reviewed sources, assembled through a documented search protocol and coded separately from the literature cited for framing, is organised along an evaluation ladder distinguishing technical performance, adoption and sustained use, and farmer outcomes. The evidence concentrates at the technical rung, where near-perfect accuracies are reported on curated data. Five experimental farmer-outcome evaluations exist, two of them South Asian; their effects range from null to modest, and none evaluates a machine-learned system, so the marginal contribution of AI over simpler digital delivery remains unmeasured. The review traces this imbalance to structural causes: connectivity and gender divides, linguistic diversity, weak integration with public extension, unresolved data governance, unfavourable unit economics, and misaligned evaluation incentives. Because these constraints characterise smallholder systems generally, the implications extend well beyond the region. The article closes with an agenda built on outcome-based evaluation standards, public data infrastructure, and the integration of AI into, rather than around, existing advisory institutions.