Agentic Artificial Intelligence in Agriculture: A Systematic Mapping Review of Reported Architectures, Applications, Challenges, and Future Directions

The review by Meghraoui and Moussaid (2026) analyzes 181 studies on agentic AI in agriculture, identifying robotics and machinery (33.1%) and irrigation/water management (19.9%) as the primary application areas. While the text confirms the review covers applications and challenges, it does not explicitly detail specific climate variability or resource efficiency outcomes, instead categorizing autonomy levels with semi-autonomous systems (48%) being the most common, followed by advisory (32%) and autonomous (20%) systems.

Key findings from the systematic mapping include:

• Architecture Trends: Multi-agent systems dominate classical architectures (72%), whereas LLM-based agents favor single-agent designs (47%).
• Geographic Distribution: Research is led by China (21.5%), India (14.9%), and the United States (12.2%).
• Technological Shifts: LLM-based agents show higher rates of reasoning (55%) and tool use (32%) compared to classical agents, which rely more on planning (35%) and collaboration (89%).
• Deployment Reality: Only 26% of LLM studies and 16% of classical studies report field-tested evidence, with 66% of LLM studies lacking specified deployment settings.

This material is a systematic mapping review of agentic AI in agriculture, focusing on how autonomous and semi-autonomous software agents are being applied to complex agricultural problems such as climate variability, input optimization, crop management, and resource efficiency. Rather than presenting a single new model or field experiment, the work synthesizes reported architectures and use cases, mapping the field’s emerging design patterns. It appears to cover the spectrum from single-agent decision-support systems to multi-agent configurations in which specialized agents coordinate across sensing, prediction, planning, and actuation layers—often in combination with IoT data, remote sensing, machine learning, robotics, or simulation tools.

A key contribution is the structured characterization of where agentic AI is being deployed, how it is architecturally organized, and what remains unresolved. The review highlights that agricultural applications are especially well suited to agentic approaches because they involve distributed, heterogeneous, and time-varying decision problems: soil moisture, weather, pest pressure, equipment availability, and market conditions all interact across spatial and temporal scales. At the same time, the review underscores major practical barriers, including data quality and availability, sensor and connectivity constraints, model uncertainty, safety and controllability, explainability, interoperability, cost, and the need for rigorous on-farm validation. In short, it helps distinguish conceptually promising agentic designs from those that are realistically deployable in operational agricultural settings.

This matters because agentic AI has the potential to move agricultural intelligence beyond static dashboards or isolated predictive models toward coordinated, adaptive decision loops that can improve resilience and efficiency under changing climate and resource conditions. For technically literate readers, the review is useful as a map of the current landscape: it identifies architectural trends, application niches, and open research gaps that future work must address—such as robust multi-agent coordination, uncertainty-aware planning, human-in-the-loop oversight, and standardized evaluation methods. Its value lies in providing a structured foundation for researchers and practitioners to build more credible, deployable, and scalable agentic systems for agriculture.