Modern agriculture is astonishingly good at producing enormous quantities of food, but it has achieved much of that efficiency by making things the same.
A supermarket does not want seventeen kinds of locally bred courgette arriving in unpredictable quantities. It wants a recognised variety that can be grown by the hectare, harvested together, packed to a specification and delivered by the pallet. The same pressures apply all the way down the chain. Large machinery prefers uniform fields, wholesalers prefer large consignments and supermarkets prefer products that can be treated as interchangeable.
The result is not literally monoculture everywhere, but the economic system strongly rewards it. Thousands of edible plant varieties exist, yet only a small fraction ever reach an ordinary supermarket.
Cheap agricultural robots could change the production side of that equation. But perhaps more importantly, AI could change the market side.
Imagine a food marketplace containing not hundreds of products but millions of individual harvests. A garden has three kilos of purple carrots ready this week. A smallholding has twenty kilos of an unusually tannic cider apple. Someone has bred a new courgette that remains firm when cooked. Another grower has a few trays of intensely sour plums suitable for the Georgian condiment t’q’emali. None of these quantities is large enough to interest a conventional supermarket.
That does not mean there is no demand for them. It means that finding the demand currently costs too much.
A sufficiently intelligent marketplace removes that problem. The person looking for sour plums does not need to know who grows them. The grower does not need to advertise nationally in the hope that the right cook happens to notice. The system already knows what is growing, what is becoming ripe and what buyers are looking for, and it can connect the two.
This turns the economics of variety on its head.
If ten growers all produce the same standard tomato, they are largely competing on price. But if one grows a very sweet yellow tomato, another a strongly acidic sauce tomato and a third a soft, intensely aromatic variety that must be eaten within a day of picking, they are no longer selling interchangeable products. Each can find the people who specifically want what they grow.
And the market can become far more precise than anything possible today. You might order 500 grams of perfect strawberries for dinner tonight, in which case appearance, ripeness and flavour matter enormously. Or you might order ten kilos of jam strawberries at any point during the next month, in which case softness and cosmetic damage barely matter at all. Someone making cider might ask for twenty kilos of bittersweet apples to balance their own harvest. A recipe could request waxy potatoes, very acidic tomatoes or berries with a particular flavour profile rather than merely asking for “potatoes”, “tomatoes” or “berries”.
The supermarket has to simplify because its shelves are finite. A digital market does not. That means unusual varieties cease to be awkward inventory and become searchable characteristics. “Nobody stocks this” matters much less when there is no central stockroom.
The effect could extend well beyond preserving old varieties. It could encourage the creation of new ones. Suppose somebody discovers that one courgette plant in their garden is unusually productive, tastes excellent and produces fruit with purple streaks. They save the seed. After a few generations the trait is stable enough that they start selling the produce locally. Buyers like it. Seed is exchanged. The marketplace accumulates information about how it performs in different soils and climates.
Perhaps it eventually becomes a recognised cultivar. Perhaps legally it spends its first few years with some magnificently bureaucratic designation such as “John Doe’s unclassified courgette no. 17”. It does not matter very much. The market can still know what it is, where it came from, how it behaves and whether people enjoy eating it.
The same could happen with apples, beans, chillies, berries, herbs and flowers. Someone might become locally known for their own dahlia cultivars. Another household might maintain obscure Georgian plum varieties. A grower with unusually wet soil could specialise in crops that thrive there rather than fighting the conditions in order to produce the same vegetables as everyone else.
This would make biodiversity economically productive rather than merely something to be conserved.
There is a useful historical analogy in traditional food markets. When I lived in Tbilisi in the 1990s, markets were full of small sellers who often had only one or two things to sell. The apple sellers sat together, so you could walk from one to the next, taste different apples and haggle about the price. There was enormous diversity, but finding the thing you wanted required physical presence, time and local knowledge.
A future AI-mediated market could reproduce that bazaar on an absurd scale. Instead of walking past twenty apple sellers, you could effectively compare every apple available within fifty kilometres. The system could know which growers you had bought from before, which flavour profiles you preferred and which fruit needed selling quickly. Sellers could specialise because specialist buyers could actually find them.
The old market supported diversity because many tiny producers happened to gather in the same place. The new one could create diversity because matching tiny producers with tiny pockets of demand would become almost free.
That is the crucial distinction. Industrial food systems tend towards uniformity because complexity is expensive. Every additional cultivar complicates harvesting, storage, purchasing, labelling and distribution. Once robots can manage plants individually and AI can manage millions of small transactions, much of that cost disappears. And once complexity is cheap, sameness loses one of its greatest economic advantages.
We tend to imagine technological progress in agriculture as ever larger farms growing ever more uniform crops with ever more sophisticated machinery. The stranger possibility is almost the reverse: extraordinarily sophisticated technology supporting millions of tiny producers, countless local specialities and a continuous stream of new varieties.
The future food system might therefore be more technologically advanced than anything we have today while looking, biologically, far less industrial.
Monoculture was partly a solution to the cost of complexity. If AI and robotics make it cheap, variety may not merely survive. It may proliferate.