How a Hokkaido Farmer Built a Farm Automation System with ChatGPT and Codex

Hiroki Tomiyasu, a former civil servant turned farmer, manages about 100 hectares of crops in Hokkaido. Faced with repetitive, error‑prone tasks such as monitoring weather, greenhouse conditions, planting calculations, and record‑keeping, he found traditional automation solutions either too expensive or too fragmented.

Instead of buying a turnkey system or hiring a custom‑software team, Tomiyasu asked ChatGPT for design ideas and used Codex to generate the necessary code. He combined inexpensive hardware (ESP32 controller, motor driver), cloud services (Cloudflare Workers, D1 database), and a LINE bot to remotely open greenhouse vents, turning a simple idea into a functional system.

Beyond vent control, AI helped him in several other farm operations:

Disease identification: photos are used to suggest possible causes of leaf anomalies.

Satellite monitoring: crop health data are overlaid on a farm map for quick assessment.

Greenhouse control: remote operation of ventilation via a mobile phone.

Group‑chat bot: consolidated temperature, status, and other metrics into a single chat interface.

Seeding statistics: automated calculation of seed requirements and area coverage.

RTK‑GPS positioning: AI assisted in understanding and assembling the positioning workflow.

Airtable database integration: linked plots, tasks, and material inventories into a unified data source.

The core insight is that AI shortens the costly path between a farmer’s need and a working solution. By describing problems in natural language, Tomiyasu let AI decompose them into modules, purchase off‑the‑shelf hardware where possible, and iteratively develop code and wiring diagrams.

This approach mirrors what many small companies and solo developers are doing: leveraging AI to turn low‑priority, niche requirements into DIY solutions without large engineering teams.

Crucially, AI’s effectiveness depends on feeding it real‑world farm data—temperature, plot conditions, equipment status, etc.—into readable formats. Tomiyasu digitized his operations by logging data in Airtable, connecting sensors to the bot, and overlaying satellite imagery, enabling AI to provide concrete assistance.

For practitioners, the article suggests a five‑step prompt to break down any on‑site problem: identify required data, pinpoint automatable steps, list low‑cost hardware or software, outline a minimum viable product, and mark where human verification remains essential.

Overall, the case shows that AI can empower individuals who deeply understand their environment to temporarily acquire engineering capabilities, turning fragmented field knowledge into systematic, reusable processes.