A Broccoli Farmer Built Greenhouse Remote Control With ChatGPT. No Engineers on Payroll.

Via besthub: How a Hokkaido farmer built a farm automation system with ChatGPT and Codex

Your messaging app can open a greenhouse now

Hiroki Tomiyasu, a farmer in Hokkaido, Japan, runs roughly 100 hectares of broccoli, pumpkins, green onions, and soybeans — and he built remote greenhouse vent control using ChatGPT, OpenAI Codex, a $30-class ESP32 microcontroller, and the LINE messaging app, according to a BestHub case study and OpenAI's profile of his work.

For owner-led businesses, that is the headline: a former civil servant with no engineering team described problems in plain language, let AI decompose them into modules, bought off-the-shelf parts, and shipped a working system. Text 「モーター 正転」 (motor forward) on LINE rolls the greenhouse vinyl up. Forward, reverse, stop — from a phone in the field.

The upside is obvious. The oversight gap is too. When one motivated operator can wire cloud workers, databases, and physical motors without a ticket queue, your governance model either keeps up or gets bypassed entirely.

Why this beats another SaaS brochure

Traditional farm automation often means proprietary machinery and specialists on retainer — priced for operations much larger than Tomiyasu's. He took a different path: ChatGPT for design conversations, Codex for code generation, ESP32 plus a motor driver for hardware, and Cloudflare Workers with a database layer for the cloud glue.

OpenAI quoted him comparing the tools to "an ultra-talented engineer always by your side." That is marketing copy, but the build log is concrete — dead motor drivers, safety checks after reboot, trial-and-error on wiring. Real IoT, not a slide deck.

He did not stop at vents. Reporting describes satellite-based crop monitoring, disease spotting from imagery, Airtable for operational logs, and a farm group-chat bot to coordinate daily work. AI here is not a chat toy; it is the shortest path from I need this on my plot tomorrow to something that runs.

The lesson for SMEs outside agriculture is identical: the bottleneck moved from "can we afford developers?" to "who verifies the thing actually works — and who owns it when it does not?"

What sensible operators copy from the broccoli field

• Start with one painful manual task. Tomiyasu began with greenhouse vents — repetitive, weather-sensitive, error-prone. Not "digital transformation." One job.
• Feed AI real operational data. Temperature logs, field notes, equipment status in readable formats (spreadsheets, Airtable, sensors). Garbage in still means garbage out — just faster.
• Mark human verification gates. His build included safety logic after stray commands on boot. Any physical automation needs an explicit "human must confirm" step.
• Document what was built. If only one person understands the ESP32 wiring and Cloudflare worker, you have a hero dependency — not an asset.
• Separate experiment from production. DIY wins in prototyping. Production needs uptime targets, rollback, and someone on call.

BestHub distilled his approach into a five-step prompt pattern: identify data, find automatable steps, list cheap hardware, sketch a minimum viable product, flag where humans stay in the loop. That framework travels well beyond Hokkaido.

DIY speed still needs an operator

Tomiyasu's story is the upside case grassroots AI stories are made of — and the reason governance cannot be optional once the builds multiply.

This is the work AgentsROI.ai does.

• A Workflow ROI Audit finds which tasks deserve custom automation versus which deserve a bought tool — before someone wires a motor from a chat window.
• Managed AI Operations tracks what actually runs in production, who built it, and what breaks when that person is on vacation.
• Staff Training and AI Adoption channels DIY energy into approved patterns — so innovation does not become shadow infrastructure.

AgentsROI.ai is not going to wire your greenhouse. It will help you decide what belongs in production, what stays in a sandbox, and who owns the outcome when the motor driver was dead on arrival (it happens).

The engineer on your phone is not a policy

ChatGPT helped a broccoli farmer roll up vinyl from LINE. That is delightful — and a governance test for every SME letting staff "just build something."

Book a Workflow ROI Audit and map where DIY AI saves real hours versus where it creates silent dependencies nobody can maintain.

This article summarizes publicly reported information from BestHub, OpenAI's ChatGPT Pro subscriber profile, and related reporting. It is for general informational purposes only. It does not constitute legal, tax, financial, investment, security, or compliance advice. AgentsROI.ai is not a law firm, accounting firm, or registered investment adviser. Facts, pricing, statistics, and product capabilities cited here reflect the sources listed at the time of writing and may change. Readers should verify current information independently and consult qualified professionals regarding obligations specific to their industry, jurisdiction, and circumstances—including applicable New York State and New York City requirements. Nothing in this article is an endorsement of any specific AI product, model, or provider. References to ChatGPT, Codex, OpenAI, LINE, Cloudflare, or ESP32 describe reported tools used in a documented case study and do not constitute product recommendations. Physical automation carries safety risks; consult qualified technicians before replicating hardware projects.