Eight months ago I wrote here about the future of food – from robot foragers bringing wild garlic to your door to decentralised swaps of apples for eggs. That vision was about what ends up on our plates. But behind the cuisine lies the soil, and if the soil fails, so does every other dream.
For half a century the trajectory of agriculture has been simple: fewer farmers, bigger machines, larger farms. Smallholders were told to modernise or sell up. Many did, and the result was a landscape dominated by giant fields and giant tractors. Economies of scale won, labour costs fell, and food production soared.
But it may not last.
The economics of scale only work when two conditions hold: the soil must behave like an obedient factory floor, and human labour must be too expensive to employ in large numbers. That bargain held for decades. Industrial agriculture traded healthy soil for predictability and cut labour to the bone with massive machines. But now the ground itself is giving way, and the cost of labour is being rewritten by automation.
At first glance, automation looks like more of the same. John Deere and others are already selling autonomous tractors. They steer themselves, work through the night, and cut labour costs further. But they are still 20-tonne machines. They still compact the soil, still spray broad-spectrum chemicals, still require the uniformity of monoculture. In truth, they are not a revolution but the logical endpoint of the old model: the same agriculture, only with no driver in the cab.
The real disruption lies elsewhere: swarms of small, lightweight bots able to treat plants as individuals. Instead of blanketing a hectare with fertiliser, they feed the one stalk that needs it. Instead of poisoning a whole field for a handful of weeds, they pull or zap them one by one. Because these bots are light, they don’t crush the soil. Because they are numerous, they make labour-intensive farming economically viable again – only now the “labour” is robotic.
The second vision is still embryonic. Vineyard robots can weed between vines, orchard bots can spray precisely, Australian prototypes are mowing paddocks. But nothing yet can manage a whole hectare as a living mosaic of plants. The technology is inching forward, not racing.
- Now–5 years: Niche robots in vineyards, orchards, greenhouses. Useful but limited.
- 5–10 years: Semi-generalist bots for small plots, able to handle mixed crops with supervision.
- 10–20 years: Field-ready swarms. Rugged, solar-assisted, repairable. Capable of mapping hectares, identifying plants individually, and working autonomously for weeks. At this point a 20-hectare bot-equipped farm could outperform a 200-hectare tractor-based one, not only in soil health but in pure economics.
Beyond that, farming becomes an ecosystem of drones, soil sensors, and ground bots, all coordinated by software.
This shift matters because it strikes at the two pillars of megafarming: labour and land. Labour costs were the reason small farms died – a family simply couldn’t weed or harvest at the same speed as a combine. But if robot labour is cheap and abundant, that advantage collapses.
At the same time, soil health is starting to show up in land values. In parts of Germany, carbon-rich soils already fetch higher prices. As carbon markets and ecosystem services expand, a hectare of fertile soil will be worth more than a compacted one. That flips the incentive: small farms with light robots can build value into their land, while megafarms strip it out.
And there’s a biological truth that economists often ignore: the more labour you invest per square metre, the higher the yield. A kitchen garden, an allotment, a well-tended vegetable plot – all outperform broadacre farming per unit of land. Megafarms only look efficient when you measure in tonnes per worker, not per metre. Robots change that metric, because they bring labour intensity back without the drudgery.
The irony is sharp. Labour costs once forced small farms out. Now robotic labour could pull them back in, hectare by hectare. Not through political revolution but through simple economics: bot-enabled smallholders paying more per hectare than their industrial neighbours. Slowly, the ownership map tilts back.
How much land does a family actually need if tended with care? Agronomists usually estimate that one to two hectares, managed intensively, is enough to feed a family of four comfortably – less if diets are plant-heavy, more if meat is central. Even a well-planned half hectare can supply most staples when labour is abundant. The constraint has never been the soil’s generosity but the cost of time and effort.
With robots, that constraint disappears. If swarms of lightweight bots can bring the labour intensity of a kitchen garden to the scale of a hectare field, then self-sufficiency stops being a romantic hobby and becomes an economic option. In twenty years’ time it may make financial sense for a family to buy a modest plot, equip it with bots, and cover their food needs while keeping their virtual office jobs. That way they are close enough to deal with any problems – whether the bots malfunction or the livestock needs attention.
And here’s the crucial point: such a farm would be run just as well as any larger unit. It would not be quaint or inefficient, but as productive per square metre as any industrial neighbour. In other words, it cannot be outcompeted – because the supposed advantages of scale vanish when labour is robotic and soil health is capital.
This dovetails with the vision I sketched earlier of robot foragers, local swaps and seasonal eating. A world of small, bot-enabled farms is also a world of resilient food networks, where families produce much of their own food and exchange the rest. It’s not just soil and land ownership that change, but the flavour of everyday life.
That would be a quiet revolution: farming as a side-economy, soil health as wealth, and self-sufficiency as rational finance. The giants would not collapse overnight, but the drift of land ownership and food production could slowly return to human scale – this time supported, not replaced, by automation.