Answers are grounded in this paper and cite their sources.
Feeding humanity at scale represents one of the greatest governance challenges of the 21st century. With the global population expected to rise to 9.1 billion by 2050, demand for agricultural products is already acute. Recent advancements in smart agriculture through agentic geospatial intelligence have shifted farmers' autonomy in agricultural activities, yet the systematic exploration of how agents perceive data, reason, and act remains underdeveloped. This review developed an agent-centric system decomposition (ACSD) and PRISMA to understand how agents engage with earth observation data, reason over it, and act on it to achieve a context-specific output in smart agriculture. Results show that 81.8% of systems operate at or below adaptive multi-agent levels (L3), with 57.6% treating geospatial data as passive inputs. None of the 33 reviewed systems demonstrated full spatiotemporal uncertainty propagation or L5 autonomy with active spatial reasoning, uncertainty-aware decision-making, and autonomous actuation. Different systems also evidenced information overload processing through edge-based data processing and execution, which can lead to challenges in effectively managing and interpreting the vast amounts of data generated in real-time applications. Descriptive synthesis showed that IoT was explicitly used as a sensing or data-acquisition source in 9 of the 33 reviewed systems (27.3%). The review identified three spatial reasoning modalities: Consumption, Inference, and Actuation, and further proposed a system for improving agents' reasoning and decision-support systems.
No takes yet. Share an insight, caveat, or question.
Ouma et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: