This paper presents FAIRY, a full‑stack smart‑agriculture agent system deployed on a soybean research farm at Harbin Institute of Technology. The system operates across the entire season, covering ridge preparation, planting, irrigation, fertilization, pest and disease treatment, harvest, grain handling, drying, and storage.
FAIRY follows an "everything is an event" execution paradigm, representing spatiotemporal evolution, remote sensing and UAV observations, sensor readings, crop‑growth transitions, machinery actions, and management interventions as state‑changing events within a shared farm process engine. This unified event model enables consistent scheduling and replay of heterogeneous data streams.
The platform integrates production‑grade machinery APIs, fixed soil and canopy sensors, multispectral and thermal drones, satellite vegetation products, a weather station, calibrated crop‑process models, agronomic records, and multi‑season yield histories. Built on the event‑driven world model, FAIRY implements a complete agentic stack: a knowledge library of atomic agronomic skills, multi‑agent controllers with orchestration backends, frontier‑ and edge‑model execution, full‑path trace logging, and deployment profiling on local nodes.
For evaluation, FAIRY benchmarks nine state‑of‑the‑art agent controllers across one hundred full‑season soybean scenarios that preserve spatial observations in a 64‑ridge field, temporal decision sequences, agronomic constraints, delayed effects, and final yield coupling. The evaluation suite combines agent success, full‑path spatiotemporal correctness, token cost, and edge‑device runtime into a unified metric set.
Results indicate that while some agents achieve high spatiotemporal fidelity, they are limited by token consumption or edge runtime constraints. The suite highlights trade‑offs between resource usage and decision accuracy, offering empirical guidance for future smart‑agriculture agent designs.
Review