Agriculture is entering an extraordinary technological moment. We can measure soils in increasingly sophisticated ways, observe crops remotely, and monitor weather, moisture, biomass and plant condition. Satellites, sensors, drones and digital platforms are producing increasingly detailed pictures of what is happening in individual fields. Artificial intelligence is beginning to turn this enormous volume of information into predictions, diagnoses and recommendations.
The technological capability to understand agriculture is accelerating, but there is a problem. What happens when the system finally knows exactly what a particular field needs? Do we have the technology that can actually deliver it? And this may turn out to be the most important unanswered question in the entire ag-tech revolution.
For years, precision agriculture has been driven by a compelling promise: if we can measure agricultural systems more precisely, we can manage them more precisely. That simply logic is correct — but incomplete.
Information has no agricultural value by itself. A soil map does not change a soil. A satellite image does not feed a crop. An AI model does not put nutrients into the root zone. At some point, information has to become physical intervention.
The real chain therefore looks like this: soil and crop condition → diagnosis → prescription → intervention → biological response → economic result → new data.
The first half of this chain is developing rapidly, but the second half is much less transformed. And this creates a dangerous paradox.
The better our measurement and analytical technologies become, the more precisely they can tell us that two neighbouring fields — or even two parts of the same field — should not necessarily receive the same treatment.
Yet much of agriculture still operates with products designed for relatively standardised application.
The soil’s early warning system: microbes can predict desertification before plants start to die
We are building increasingly intelligent agricultural systems while retaining comparatively blunt instruments for acting on their intelligence. And that is the current bottleneck!
That matters because ag-tech is not free. Sensors and data platforms cost money. Remote sensing, analytics, AI infrastructure, sampling, connectivity and integration all require investment.
But the economic justification for this infrastructure ultimately depends on one thing: Does better information create enough additional value in the field to pay for itself?
If the answer is only a slightly better recommendation applied through essentially the same standardised input system, the economic upside may remain limited. And this is where the agricultural data revolution could hit a wall. Not a wall of insufficient information, but a wall of insufficient ability to act on what the data tell us.
Therefore, the next agricultural breakthrough cannot be another sensor or another layer of data collection. Perhaps we should already start asking: What kind of agricultural input would be required if we already knew, with increasing confidence, exactly what an individual field needed?
Not simply “How much fertiliser should this field receive?” But which unique formulation with which nutrient balance, which dose, at what moment, under which climatic conditions, applied in which way, and how should the prescription change as the field changes?
That is a fundamentally different concept, and it means moving from precision agriculture towards adaptive agriculture. And the unit of treatment changes with it.
It is no longer the region, the farm or even the crop. The unit of agricultural treatment becomes the individual field. And potentially, over time, the changing state of that field.
This is where the technological challenge becomes much more interesting. A genuinely adaptive system requires the ability to translate a digital prescription into a physical product or intervention with sufficient flexibility. The more sophisticated the diagnosis becomes, the greater the required freedom of intervention.
This is not simply a matter of producing “better fertiliser”. It is a matter of creating an input architecture capable of responding to increasingly precise soil intelligence.
Traditional fertiliser systems were largely designed around producing standardised products efficiently at scale. The new model could be almost the inverse: standardised building blocks upstream and personalised intervention downstream.
The agricultural input industry could therefore evolve towards something resembling mass customisation: highly efficient production of configurable components, followed by field-specific formulation, dosage, timing and delivery. That is what would allow data to become action.
Artificial intelligence will undoubtedly become increasingly important.
It can integrate soil data, weather, crop development, field history and previous interventions. It can identify patterns that are impossible for a human agronomist to process at scale. And it can generate increasingly sophisticated recommendations.
But AI has a fundamental limitation. It cannot physically intervene in the field.
The truly powerful farming system therefore looks more like a closed biological learning loop: observe → understand → prescribe → intervene → measure → learn → intervene again.
Every field becomes a learning environment. Every season generates new evidence. Every intervention improves the knowledge available for the next intervention. And the value of the entire system compounds over time.
This leads to a provocative conclusion. Perhaps the agricultural technology revolution should not be judged primarily by how much agricultural data we can collect, but by how much better we can act because we have collected it.
If the industry develops increasingly sophisticated systems for observing fields but does not develop sufficiently flexible systems for intervening in them, we may end up with an extraordinary technological infrastructure whose economic return is frustratingly modest. And that would be a remarkable outcome: agriculture becoming vastly more measurable without becoming proportionally more controllable.
The real breakthrough therefore lies in closing the loop. Data must become prescription, prescription must become intervention, intervention must become measurable outcome. And outcome must become learning. Only then does the data revolution become a productivity revolution.
This is also why the next generation of agricultural input companies may look very different from today's.
The winning capability may not be the ability to manufacture the largest volume of a particular fertiliser. But it may be the ability to manufacture the right intervention for the right field at the right moment — repeatedly, economically and at scale.
If that capability can be developed, the relationship between digital agriculture and agricultural inputs changes completely.
The input is no longer merely a commodity being recommended by a digital system. It becomes the physical execution layer of that digital system. And that is potentially the missing piece of the entire ag-tech architecture.
The question facing agriculture is therefore no longer simply: “Can we know what this field needs?” Increasingly, the answer will be yes. The much more important question is: “Can we actually give the field what we know it needs?”
If the answer is no, then the ag-tech revolution remains incomplete. If the answer becomes yes, a new agricultural model becomes possible. One in which every field can receive increasingly personalised fertilisation, continuously adapted to its soil, crop, climate, history and economic objectives.
That is not simply better fertilisation. It is a different relationship between information and agriculture. And perhaps that is where the real agricultural revolution finally begins.
