There's a company I keep thinking about. I'll be in the middle of something — making tea, staring at a half-finished CAD model — and my brain just drifts back to them.

The company is eternal.ag. They build fully autonomous harvesting robots for greenhouses. Their tagline is "building resilience into food production." I applied for their product designer role.

I'm writing this anyway.

video sourced from eternal.ag official website

What They're Building

What they're doing is, on the surface, pretty legible: greenhouse labour in Europe has dropped by nearly 30% since 2010. Someone has to pick the tomatoes. If humans increasingly won't — or can't — a robot has to. eternal.ag's answer is Harvester, a fully autonomous robot that runs up to 22 hours a day, uses AI to identify and pick produce, and feeds every action back into a system that keeps getting better.

They train and validate in virtual greenhouses first — cutting iteration cycles from months to days — then deploy into the real thing. Every deployed robot becomes a data point. The system learns. It scales. By 2040, they want no manual labour in greenhouses at all.

The Tagline That Won't Leave Me Alone

"Building resilience into food production." It's not a tagline. It's a design brief.

What does resilience look like as a physical object? How do you design a machine that works reliably in a greenhouse — humid, organic, variable — and keeps working even when things go wrong? How does it feel to a grower who has to trust it completely, day after day, without babysitting it?

These are the questions I find myself sketching around. Not because I was asked to, but because the problem is genuinely fascinating.

Why I Think I'm the Right Kind of Wrong Fit

My grad project was building a 3D printer from scratch. Not designing one — building one, from a BOM I wrote myself, with components I sourced for cost, with tolerances I had to actually hit. It taught me that a machine is a negotiation: between what you want it to do, what materials and physics will allow, and what a real person can actually live with. You can't render your way out of a bad kinematic decision.

A harvesting robot is the same negotiation, at higher stakes. The crop doesn't cooperate. The greenhouse layout varies. Every design decision — the gripper geometry, the sensor placement, the way the machine communicates its status — has to work in the real world. Not in a render. Not in a virtual greenhouse. In the actual one, at 5am, when no one is watching.

The Part I Keep Coming Back To

eternal.ag's platform is modular. The plan isn't just to harvest tomatoes — it's to build something that can expand to other crops, other tasks, other greenhouses. That's a systems design problem as much as a product design problem. And systems design is the kind of work where getting the physical and digital layers to talk to each other — really talk, not just coexist — is the hardest thing to get right.

Closer to Home Than I Expected

Today I was working with copra. My family runs a small business converting coconuts to copra — they've been doing it for years. I grew up watching it happen, and somewhere in the middle of today I found myself thinking: why isn't there a machine for this?

My dad did invest in one small step forward — a dryer, an oven-like machine that dries the copra instead of sun drying it. When I was smaller, we used to lay them out one by one under the sun. At least a thousand pieces a day, placed in a straight line. It was the kind of work that is hard to explain to someone who hasn't done it — repetitive, physical, unrelenting. The dryer changed that. One machine, one decision, and a whole layer of labour just disappeared.

But the rest of it is still manual. The cracking, the extraction — getting the copra and the pongu (the white edible sprout that grows inside a mature coconut) out cleanly, without damage, at any kind of scale. Nobody has made a machine for that yet, as far as I know. And I've been thinking about why. The coconut is variable. The interior geometry changes. The sprout is delicate. These are exactly the kinds of problems that make automation hard — and exactly the kinds of problems eternal.ag is learning to solve for tomatoes.

I don't know if they'll come back to me. Probably not. But I find myself thinking about the grower's experience, about what it feels like to hand trust over to a machine, about what the robot does when something goes wrong and a human needs to step in. I'm thinking about it because the problem is real — food security, climate pressure, the slow erosion of agricultural labour — and eternal.ag is one of the few companies trying to design its way through those problems, not around them.

And maybe I'm drawn to it because I've seen what even one small machine can change. What a thousand coconuts a day looks like before and after someone decides to solve it.

That's worth obsessing over, even if it doesn't work out.