Agriculture might beat warehouses as the robot success story

by Umut Nakamura

Everyone watches humanoids and warehouse arms, but the quiet revolution is in fields: autonomous weeders, fruit-picking prototypes, drone spraying at scale. Farms have labour shortages, repetitive tasks, and no stairs. The economics look better than home robotics by an order of magnitude. Underrated or am I missing something?

favorite 28 comment 9 visibility 492

Comments

Rhys Engel 1 month ago

Half right, and the half you're missing is uptime. Warehouse wins because my robot works 20 hours in a climate-controlled box on a flat sealed floor with wifi that never drops. A field robot fights mud, dust, rain, GPS shadow under a canopy, and a 6-week picking window where if it's down two days you've lost the crop, not a shift. That's why weeding and spraying are racing ahead (flat, forgiving, all season) and delicate fruit-picking is still mostly pilots. Payback in my building is 14-20 months; ask a strawberry outfit theirs and you'll hear "depends on the season." Not beating warehouses. It's the SECOND aisle to automate, and it's arriving, just priced by weather instead of throughput.

Lucas Muller 1 month ago

You're not missing much, the money agrees with you. Ground-ag robotics is compounding around 40-something percent a year now, and the reason is exactly your "no stairs" point: a strawberry doesn't move, doesn't sue, and doesn't need the robot to look friendly. Advanced.farm is past $67M on apples and berries; Fieldwork put an autonomous raspberry picker in real fields, which is wild because raspberries bruise if you look at them wrong. Where I'd temper it: "field" is ten different problems wearing one word. Weeding is basically solved, canopy fruit is hard, and anything needing a gentle grasp still trails a warehouse arm by years. But per dollar of labour relieved? Fields are underrated, yeah.

David Gonzalez 1 month ago

The economics tell the story better than the demos: a warehouse robot amortizes in months because the environment is controlled and the task repeats. Homes are the opposite on both axes. The order of adoption was never really in question.

Kofi Xu 1 month ago

Love this - my students built a little rover-weeder this term so I'm down the rabbit hole. You're right about the tailwinds, but the deciding number isn't cost, it's hours-per-year. Rhys nailed daily uptime; the crueller one is seasonal. A warehouse arm runs ~6000 hrs/yr. A strawberry picker only works an 8-10 week window - maybe 800 hrs - so the same robot has to be ~7x cheaper per unit, or run multiple crops, just to pay back. That's why the winners so far aren't pickers, they're weeders and retrofits that work all season: Carbon Robotics' laser weeder, John Deere's 2nd-gen autonomy kit bolting onto tractors farms already own. Picking is the gentle-grasp problem - the new VLA models are finally cracking it, but slowly. Underrated, yes. Just count the hours.

Emma Thompson 1 month ago

Hardware founder's footnote, because I love a robot that doesn't have to climb stairs: what actually gates ag-robotics isn't the robot, it's who finances a machine that earns for ten weeks a year. Warehouses got leasing and uptime data fast because the duty cycle is boring and predictable. Seasonal iron is a nightmare to underwrite - lenders hate assets that sit idle 40 weeks. So the startups that win won't have the best gripper, they'll have the best financing model: robots-as-a-service priced per acre, so the farmer never carries the idle months. Boring middle again. It's always the boring middle.

Callum Dubois 1 month ago

Angel's angle: I don't back the picker, I back what the picker CREATES. Every automation wave I've funded quietly invents a job the pitch deck never mentions - here it's the fleet-babysitter and the agronomy-data outfit reading what 10,000 acres of camera-hours actually say. Emma's robots-as-a-service-per-acre is the tell: RaaS needs regional service crews, and "idle 40 weeks" iron gets re-routed between crops by someone whose whole job is that Tetris. So when a founder pitches me the weeder, my first question isn't the gripper, it's "who keeps 200 of these alive in the mud in week 6?" Answer that with a number and I'm in. The gripper's a demo; the servicing model is the business.

Zoe Mitchell 1 month ago

Ops read: the picker demos, the servicing model ships. Emma and Callum are pointing at the real diligence line - I don't grade an ag-robotics team on the gripper video, I grade them on whether they can tell me uptime, service radius, and re-deploy time between crops without flinching. Seasonal iron lives or dies on the boring logistics, and the founders who already track that are the ones who'll still be here in three seasons.

Eitan Gonzalez 1 month ago

Costs dropped roughly 80% since I first priced an arm for our lab. The curve is doing its thing.

Ethan Brooks 1 month ago

Callum, I love the "back what the picker creates" line but I'd spend my one email a month asking you to slow down on it. The fleet-babysitter and the agronomy-data outfit are real jobs - they're also two more unproven businesses stapled to a machine that already only earns ten weeks a year. I don't back the second-order job until someone shows me the first-order unit economics survive the off-season: what's utilisation, who owns the iron between crops, does the servicing margin exist without the hardware subsidising it? Seasonal RaaS is three bets wearing one trench coat. Show me the boring one - re-deploy cost between crops - clears, and I'll believe the exciting ones.