Agriculture and Artificial Intelligence: Quick Gains in Forecasting and Targeted Inputs, a Long Road from Lab to Field

Purpose. Farming is carried out outdoors, on millions of holdings of very different size, and its main risks come from weather and biology. This article asks where artificial intelligence stands in agriculture in 2026 and where it could be by about 2035. Design and methods. A review combining the share of the agricultural literature that involves AI (OpenAlex, 2010–2025) with peer-reviewed studies of prediction, crop-disease recognition and weather forecasting, field-trial reports on targeted spraying, a regulated evaluation of an AI advisory service, and statistical surveys of technology use on farms. Claims taken from the literature were checked against the sources cited. The scenarios are judgements built on this evidence. Findings. The share of agricultural works that mention AI rose from 1.1% in 2018 to 10.1% in 2025, 9.5 times as much, the third-largest increase of the 20 industries we compared; language models are only 3.2% of AI-related works, so the field is still mostly about vision and prediction. Three results stand out. Random-forest models predicted crop yields with errors of 6–14% of the mean against 14–49% for linear regression. A plant-disease classifier that reached 99.35% accuracy on laboratory-style images fell to about 31% on images taken under different conditions. And machine-learning weather models have begun to match physics-based systems: one outperformed the best operational deterministic forecast on 90% of 1,380 verification targets, and the European centre put an AI forecast into operations in 2025. Targeted spraying reduced herbicide use by 43–59% in a three-year soybean trial, but its weed seed bank result at low sensitivity is a warning. Randomised evidence on AI advice for smallholders is still pending. We expect forecasting and targeted-input tools to be routine by 2028, evidence on advisory services and farm robotics to arrive between 2027 and 2032, and the largest uncertainty to be how the benefits reach small farms. Originality. The article sets the strongest evidence for AI in agriculture beside the gap between laboratory and field performance, uses measures that apply to a sector with little adoption statistics, and flags which expectations rest on trials that have not yet reported.