To reclaim his time, Paul constructed a neighborhood, multi-agent AI system utilizing Gemini 3.6 Flash inside Google Antigravity to combine siloed farm information and calculate day by day profitability.
As a substitute of APIs or net scraping, the system makes use of a neighborhood, directory-based file-interface. When CSV information exports or pictures of papers receipts, PDFs, and invoices are saved to a monitored folder, Gemini’s multimodal energy extracts and merges visible and numeric metrics — holding information beneath Paul’s management.
A specialised multi-agent workflow replaces lengthy prompts with specialised, orchestrated roles:
- Orchestrator: Manages the general day by day workflow.
- Ingestion Brokers: Standardize uncooked recordsdata (milking robotic exports, feed logs).
- Evaluation Agent: Evaluates organic and climate impacts.
- Reporting Agent: Generates clear, pure language summaries.
The system routinely transforms uncooked recordsdata right into a cohesive enterprise overview.
Leveling the taking part in subject for small companies
Finally, Paul’s ambition is to empower unbiased farmers with know-how they’ll use to streamline their operations the identical means he has been capable of for his personal enterprise. Not solely is constructing the instrument a hurdle for a lot of, however as soon as they’ve them, working at farm scale was a value problem that wanted to be solved.
The day by day agentic workflows that require steady reasoning, parsing, and power execution have been too costly for small companies like his. Gemini 3.6 Flash helped make Paul’s day by day operation cheaper.
Designed for superior reasoning, instrument use, and coding, Gemini 3.6 Flash incorporates a 1 million token context window and a 64,000 token most output. Benchmark evaluations present it achieves roughly a 17 % discount in output tokens in comparison with Gemini 3.5 Flash, at a decrease value per output token, considerably decreasing the price of working agentic loops.
Measuring success by Static Variable Margin
The system is designed to optimize for Paul’s major metric: Every day Static Variable Margin (SVM). Not like conventional metrics like Revenue Over Feed Price, which fluctuates with risky milk and feed costs, SVM holds market costs fixed, isolating the true organic and operational effectivity from market noise. This grounds the agentic system in a fact Paul can depend on to offer him actionable insights for farm operations.
