Agricultural Horizons: Research, Technology and Sustainable Development Vol. 5

Artificial intelligence (AI) is moving agricultural extension beyond digitised information delivery towards systems that can generate, retrieve, translate, personalise and synthesise advisory content in conversational form. This shift is consequential for extension professionals because generative AI can perform portions of knowledge work that were previously inseparable from professional practice, while also introducing new risks of inaccurate, decontextualised or unaccountable advice. This critical narrative review examines how AI, particularly large language models (LLMs), retrieval-augmented generation and related conversational systems, is reshaping the work of agricultural extension professionals. Literature published primarily from 2015 to 2 August 2026 was identified through multidisciplinary, agricultural, educational and computing scholarly sources, with earlier foundational studies retained where conceptually necessary. The synthesis distinguishes technical capability from demonstrated advisory effectiveness and from downstream farmer outcomes. Evidence indicates that generative AI can support drafting, summarisation, translation, question answering, first-pass diagnosis, programme planning and administrative work, while domain grounding and expert review can improve relevance. Yet current agricultural evidence remains dominated by prototype evaluations, expert ratings, cross-sectional surveys and short-term implementation studies rather than longitudinal studies of professional performance or farmer welfare. Site-specific agronomic errors, hallucination, language and equity limitations, automation bias, privacy concerns and unclear liability constrain direct substitution of professional judgement. The emerging role of the extension professional is therefore better understood as a human–AI stewardship role that combines domain expertise with evidence curation, contextual validation, risk triage, relationship building, accountability and participatory governance. Professional development must correspondingly extend beyond prompt use to include AI literacy, source verification, calibrated reliance, data governance, equity assessment and evaluation of field outcomes. Extension organisations should adopt risk-tiered oversight, locally grounded knowledge systems, auditable workflows and explicit escalation rules. The central challenge is not whether AI can generate agricultural advice, but how extension systems can use it without weakening the contextual expertise, trust and public accountability on which credible advisory services depend.