A Japanese farmer manages 250 acres with OpenAI's Codex

Modernizing a 250-acre farm usually requires either a massive tech budget or an expensive enterprise agriculture contract.

A farmer in Hokkaido just bypassed both by having AI write his entire infrastructure from scratch.

OpenAI highlighted Hiroki Tomiyasu, a former civil servant who took up farming without formal agricultural training. He now manages 100 hectares of farmland across Hokkaido, growing broccoli, pumpkins, green onions, and soybeans. Despite having zero engineering background, Tomiyasu built a sophisticated farm automation setup entirely powered by ChatGPT and Codex.

From microcontrollers to orbit

Tomiyasu currently runs 8 active AI-powered workflows managing his daily operations.

At ground level, he uses ESP32 microcontrollers to handle automated motor controls while feeding telemetry from field sensors into unified dashboards to track greenhouse temperatures.

Looking down from above, he automated pipelines that pull satellite imagery to generate NDVI vegetation maps, tracking crop health across his plots in near real time.

Codex generated the glue code connecting the IoT hardware, cloud data, and satellite APIs. Tomiyasu, who documents his builds on note.com, describes the workflow as having an ultra-talented software engineer sitting right next to him at all times.

Why it matters

We think this story exposes a massive shift in how software gets built. For decades, the tech industry assumed technical skill was the gating factor to building useful tools. Tomiyasu proved that once the syntax barrier is removed, deep domain knowledge is what actually matters.

In specialized industries like agriculture or manufacturing, the hardest part of software was never writing an API call. It was translating field experience into functional logic. Outside developers rarely understand soil moisture patterns or greenhouse venting schedules, while field operators could not code.

Now, domain experts who know exactly what needs fixing can generate working systems on their own. While physical realities like broken sensors and hardware wiring still require manual grit, the code itself is no longer the bottleneck holding builders back.