Agricultural Horizons: Research, Technology and Sustainable Development Vol. 6

Agricultural extension is increasingly expected to deliver advice that is not merely digital, but timely, farm-specific, agronomically defensible and responsive to farmers' changing constraints. This expectation has accelerated interest in artificial intelligence (AI), farmer-generated data and context-aware advisory systems. This critical narrative review evaluates whether these technologies are strengthening agricultural extension in ways that can be distinguished from the effects of digital delivery or personalisation alone. The review integrates evidence on mobile and site-specific advisory services, decision-support tools, machine learning, generative AI, retrieval-augmented generation, farmer data governance, digital inclusion and human-AI collaboration. The strongest causal evidence concerns personalisation and site-specific agronomic information rather than AI itself. Randomised trials show that tailored nutrient and management recommendations can improve adoption and, in some settings, yields, profits or resource-use efficiency, but other trials find behavioural change without measurable productivity gains. Evidence specific to generative AI remains much less mature and is dominated by expert-rating studies, developer-led evaluations, observational studies and conceptual architectures. Large language models can expand conversational access and synthesise heterogeneous knowledge, yet their most consequential errors occur precisely where extension requires local specificity, including planting dates, input rates and timing. Personalisation therefore depends less on model sophistication alone than on the quality, provenance and granularity of farmer, field, weather and agronomic data; explicit decision rules; accountable human oversight; and institutions that sustain trust. Data governance is not an ancillary ethical issue because consent, control, interoperability, representation and benefit sharing influence both data quality and farmer participation. The review concludes that AI should be treated as one layer within a socio-technical extension system rather than as a substitute for agronomy or extension professionals. Future evaluation should test the incremental value of AI against simpler personalised digital services, measure distributional as well as average effects, and combine agronomic safety benchmarks with longitudinal farm-level outcomes.