Economics, Business and Management: Recent Advances Vol. 5

Generative artificial intelligence (GenAI), particularly large language models (LLMs), is entering agricultural extension at a point when conventional advisory systems already combine face-to-face support, mobile services, digital platforms, sensors and data-driven decision tools. This critical narrative review evaluates whether GenAI changes the capabilities, risks and institutional logic of extension rather than simply adding another communication channel. Literature was selected through live searches of multidisciplinary, agricultural, educational and biomedical scholarly sources, supplemented by citation searching and bibliographic verification. The synthesis focuses on farmer-facing and adviser-facing conversational systems, retrieval-augmented generation, localisation through expert feedback, multimodal and tool-using systems, and integrations with weather, sensor and geospatial data. Emerging studies show that LLMs can generate useful agronomic explanations, improve access to technical material and support rapid question answering, while domain grounding and human review can improve contextual relevance. Evidence is substantially weaker for sustained changes in farm decisions, productivity, profitability, resource efficiency, safety and distributional outcomes. Performance also varies with crop, location, question type, model version, prompt design, knowledge source and evaluator expertise. Consequently, benchmark accuracy or favourable expert ratings should not be treated as proof of effective extension. The principal challenge is to embed generative models within accountable advisory institutions that can verify sources, recognise uncertainty, protect farm data, escalate high-risk questions and preserve farmers' capacity to judge and adapt recommendations. The review proposes an evidence-to-action framework linking authoritative knowledge and live local data to grounded generation, safety controls, human mediation, farmer interaction and outcome monitoring. Future research should prioritise pragmatic comparative field trials, multilingual and low-literacy interfaces, reproducible model-version evaluation, safety endpoints, equity analysis, cost-effectiveness and governance. GenAI can expand the reach and responsiveness of agricultural advisory services, but its value depends on calibrated autonomy, contextual grounding and institutional responsibility rather than conversational fluency alone.