Four Gemini Flash Agents Now Help Run a Michigan Dairy Farm's Operations

Google has published a customer story about Paul Windemuller, a Michigan dairy farmer who now runs his operation with the help of a four-agent system built on Gemini Flash. The setup replaces hours of manual spreadsheet work each morning with automated data integration, freeing him up to focus on actual farm management decisions.

Details

• The farm: Windemuller runs Dream Winds Dairy, which grew from just 30 leased cows in 2014 to a herd of 260 Holsteins today. He is a 2024 Nuffield International Farming Scholar studying agricultural technology and AI
• The problem it solves: Data from milking robots, feed logs, weather stations, and various online portals used to live in silos, requiring hours each morning of downloading files, merging spreadsheets, and running calculations before any real farm work could begin
• The four-agent architecture:
  • Orchestrator: oversees the overall daily workflow
  • Ingestion Agents: standardize raw files from milking robots and feed logs
  • Analysis Agent: evaluates biological and weather-related impacts
  • Reporting Agent: produces natural-language summary reports
• A custom metric — Daily Static Variable Margin (SVM): isolates true operational efficiency from market price swings. One example output, a “Farm CEO Briefing,” pinpointed a $0.15-per-cow drop in SVM and traced it to reduced feed intake caused by rising humidity
• Model performance: Gemini 3.6 Flash cut output tokens by roughly 17% compared to earlier versions, making it economically viable to run agents continuously even on a small farm’s budget

What happened next

With data consolidation now automated, Windemuller has been able to redirect the time he used to spend on manual reporting toward herd health and broader operating strategy. Google frames the story as an example of agentic AI moving beyond office work and into hands-on, traditional industries like dairy farming.