Hands for the harvest

Thesis

Specialty crops such as berries, apples and vegetables still depend on human hands at harvest, and labor is getting scarcer and more expensive every season. Growers now rely heavily on the H-2A visa program, with several hundred thousand positions certified each year TODO: verify DOL figures. Robots that pick reliably across crops would turn the industry’s biggest variable cost into a predictable service.

Why now

• Robot foundation models are starting to generalize manipulation across objects. That’s the piece that killed earlier single-crop picking robots.
• Cheaper actuators, cameras and edge compute.
• Labor scarcity is structural, not cyclical.

Why it’s big

Harvest labor is the largest single cost for many specialty-crop growers. A harvest-as-a-service model (price per pound, no capex for the grower) removes the adoption barrier and builds a data advantage across farms.

What has to be true

1. Manipulation transfers across crops with modest fine-tuning.
2. Unit economics work at realistic uptime (dust, rain, night).
3. Growers trust a robot with a perishable crop during the most important two weeks of their year.

Biggest risk

Technical risk is real, which is the right kind. It can be resolved in field trials, not by waiting for the market. But many well-funded attempts have failed, so I’d want to know exactly what’s different.

Why I care

Automating bookkeeping at FarmerCPA taught me where automation pays off most: expensive, repetitive work that needs little judgment. In agriculture, harvest labor is the biggest example.