From fields to supply chains: AI agents drive next wave of agricultural innovation

A new review study highlights how intelligent agent systems, powered by large models and multimodal data, are emerging as a foundational pillar in the transformation toward precision, scalable, and adaptive agriculture.

Published in Electronics, the study titled “Agent Technology for Agricultural Intelligence: Methodological Framework and Applications” presents a framework that integrates core technologies, real-world agricultural applications, and future development pathways. The research outlines how agent-based systems are evolving from traditional rule-based simulations into advanced, learning-driven architectures capable of autonomous perception, decision-making, and execution across complex agricultural environments.

A new technological backbone: how intelligent agents are redefining agricultural systems

The three-layer methodological framework connects core technologies, application scenarios, and future evolution pathways. It is built around the idea that intelligent agents function as autonomous decision-making units capable of perceiving their environment, processing complex data, and executing actions in real time.

Unlike traditional agricultural models that rely on static assumptions and limited datasets, modern agent systems integrate multimodal data sources, including visual inputs from drones, sensor data from fields, and environmental information such as weather and soil conditions. This allows for a more dynamic and accurate understanding of agricultural systems.

The study identifies five key technological pillars that define the architecture of agricultural intelligent agents. These include multimodal data perception and fusion, scenario-oriented knowledge modeling with dynamic memory, intelligent decision-making and planning, embodied artificial intelligence, and closed-loop feedback optimization. Together, these components enable agents to operate in a continuous loop of perception, reasoning, action, and learning.

Multimodal data fusion is particularly significant, as it allows agents to combine heterogeneous data types into unified insights. This capability is essential in agriculture, where decision-making depends on a wide range of variables that are often difficult to integrate using traditional methods.

Knowledge modeling and memory systems further enhance agent capabilities by allowing them to store, retrieve, and update information dynamically. This reduces reliance on static datasets and enables continuous learning, although the study notes ongoing challenges related to model accuracy and hallucination in large language models.

Decision-making and planning systems, supported by frameworks such as reasoning-action integration models, allow agents to process multi-step tasks and adapt strategies in real time. Meanwhile, embodied AI bridges the gap between digital intelligence and physical execution, enabling agents to interact with real-world environments through agricultural machinery and robotics.

Closed-loop feedback optimization ensures that agents continuously improve their performance by learning from outcomes and adjusting their strategies accordingly. This iterative process transforms agents from passive tools into adaptive systems capable of self-improvement.

From crop fields to supply chains: real-world applications across agriculture

The study provides a detailed mapping of how agent technologies are being applied across four major agricultural domains: crop cultivation, resource utilization, equipment upgrading, and full industrial chain governance.

In crop cultivation, agent systems are enabling more precise and adaptive farming practices. By integrating data from sensors, drones, and environmental monitoring systems, agents can detect diseases, simulate crop growth, and optimize planting strategies. Traditional models such as APSIM and DSSAT are being enhanced with multi-agent systems to better capture the complexity of real-world agricultural decision-making.

Advanced systems now incorporate deep learning and reinforcement learning to improve disease detection and adaptability. For example, agent-driven frameworks can analyze aerial imagery to identify crop health issues and adjust interventions in real time. However, the study notes that challenges remain in processing noisy or low-resolution data and in generalizing models across diverse environments.

In resource utilization, agent technologies are playing a critical role in optimizing the allocation of water, energy, and other inputs. Integrated models combining system dynamics and agent-based approaches have demonstrated significant improvements in water allocation efficiency and reductions in energy consumption. These systems enable more sustainable agricultural practices by balancing competing demands and optimizing resource use.

The study also highlights applications in low-carbon agriculture, where agent-based models are used to evaluate renewable energy adoption and carbon reduction strategies. These approaches provide valuable insights into how agricultural systems can transition toward more sustainable practices while maintaining productivity.

In the domain of agricultural equipment, intelligent agents are driving the development of autonomous and collaborative machinery systems. Multi-agent reinforcement learning is being used to optimize tasks such as fertilization, path planning, and drone-assisted operations. These systems enable multiple machines to coordinate their actions, improving efficiency and reducing operational costs.

Digital twin technologies and high-fidelity simulations are further enhancing these capabilities by allowing agents to model and predict real-world scenarios before executing actions. This is particularly important in complex and dynamic environments where traditional decision-making approaches fall short.

At the industrial chain level, agent technologies are enabling more integrated and responsive supply chains. Agents can simulate the behavior of different stakeholders, including farmers, processors, and distributors, to optimize logistics, reduce disruptions, and improve overall system resilience.

Applications in this area include intelligent scheduling of harvesting and transportation, real-time data sharing through IoT systems, and policy simulation models that support decision-making at the governance level. These systems help transform fragmented agricultural processes into coordinated, data-driven ecosystems.

Barriers to adoption: data, scalability, and digital literacy challenges

Despite the significant potential of agent technologies, the study identifies several critical challenges that must be addressed to enable widespread adoption.

One of the most pressing issues is the limitation of agricultural data. The study highlights the challenges of integrating heterogeneous data sources, dealing with sensor noise, and ensuring data accuracy. In many cases, data collected from agricultural environments is incomplete, inconsistent, or affected by environmental factors, making it difficult for agents to generate reliable insights.

Data security and privacy also emerge as key concerns. Sensitive information, such as farming strategies and soil conditions, is at risk of exposure during data collection and sharing processes. Addressing these issues will require the development of standardized data frameworks and robust security mechanisms.

Another major challenge is the scalability and adaptability of agent-based models. Many existing systems are designed for specific scenarios and struggle to generalize across different environments or accommodate the dynamic nature of agricultural systems. Simplified assumptions in models can lead to discrepancies between simulated outcomes and real-world conditions, particularly in complex scenarios involving multiple stakeholders and variables.

The study calls for the development of more flexible and scalable architectures that can adapt to diverse agricultural contexts. Technologies such as cloud computing, federated learning, and metaverse-based simulations are identified as potential solutions for enhancing model scalability and realism.

Digital literacy among farmers represents another significant barrier. In many regions, farmers lack the technical skills and resources needed to adopt advanced technologies. This creates a gap between technological innovation and practical implementation, limiting the impact of agent-based systems.

To address this issue, the study suggests the development of user-friendly interfaces and AI-powered advisory systems that can provide real-time guidance to farmers. Collaboration between universities, enterprises, and agricultural communities is also essential to ensure that technological solutions are aligned with real-world needs.