Information system running a 15,000-tree orchard, a 60-head dairy, and a 200-head cattle operation. Structured data, live dashboards, and an agent that answers complex questions in plain language. Built for a field team with no formal education and intermittent connectivity.

Two agents for a running coach agency. One builds complex training plans and upskills junior coaches. The other pressure-tests plans against 25+ conditional rules the coaches had only ever held in their heads. Claude Cowork with a Google Sheets backend, scaled down deliberately to what two non-technical users could operate.

Two agent teams. The first (6 orchestrated sub-agents) scouts problems and validates market opportunities. The second (7 sub-agents) takes the brief and builds a working prototype to test with real users. Full-auto goes from opportunity space to prototype in about two days. I also built a human-in-the-loop version, because the autonomous one makes rational decisions I would not make.

Took farm data, productivity metrics, soil lab results, and available budget, and returned a prioritized plan for sustainable practices with the business case behind each one. 20 leads over 12 weeks, 5 assessments run, 1 farm acted. The output was well received, but farmers were reluctant to act because they were too ambitious (value prop learning that triggered a pivot).