How Gemini Flash agents are helping a Michigan dairy farmer

Paul Windemuller, a dairy farmer from Michigan, employs Gemini 3.6 Flash to automate the complex data analysis on his farm. This custom multi-agent system consolidates disparate data, such as milk quality and feed logs, to determine biological efficiency and provide daily actionable insights. This streamlined approach aids in the management of intricate biological and environmental factors prevalent in dairy farming, a sector characterized by tight profit margins.

In 2014, Paul and his wife Brittany established Dream Winds Dairy, commencing with 30 leased cows, which expanded to a modern facility housing 260 Holstein cows. As a 2024 Nuffield International Farming Scholar, Paul perceives dairy farming as a technological puzzle, leveraging continuous data streams generated by sensors, weather stations, and online portals.

However, these systems were previously segregated, prompting Paul to devise a local multi-agent AI system utilizing Gemini 3.6 Flash for seamless data integration and profit calculation. Instead of relying on APIs or web scraping, the system utilizes a directory-based file interface, enabling Gemini's multimodal capabilities to extract and amalgamate visual and numeric data from CSV exports and visualized receipts.

A specialized multi-agent workflow converts raw files into a consolidated business overview, augmenting Paul's objective to equip independent farmers with tools to optimize their operations, akin to his experience. Though creating such a tool posed significant challenges, managing it at scale presented another cost hurdle for small businesses.

Gemini 3.6 Flash addressed these concerns by offering advanced reasoning, tool use, and coding functionalities, boasting a 1-million-token context window with a 64,000 token maximum output. Its performance is superior to Gemini 3.5 Flash, exhibiting a 17% reduction in output tokens at a lower cost per output token, thereby diminishing the expense of running agentic loops.

The system aims to optimize Paul’s primary metric: Daily Static Variable Margin (SVM), which disregards market price volatility to isolate true biological and operational efficiency. To save time, the reporting agent compiles results into a daily Farm CEO Briefing, highlighting daily margin drivers and offering targeted recommendations, such as adjusting ventilation for heat stress, thereby enabling Paul to concentrate on expanding his business rather than manual spreadsheet work.