An Agricultural Revolution in Industry
2026-07-29
There is a mystery in creation. Something new comes into being where there was nothing. I was captivated by that moment, and until recently I believed that making things was everything. I only saw the value of bringing things into the world, and I gave almost no thought to what happens after. I believed development was the hardest part. It is a common affliction among people who have done nothing but development.
Nearly two years ago, I wrote:
What became clear in the process was a major challenge: the disconnect between the physical and the digital. And the key to truly transforming society lies in the physical layer, that is, on "the production floor."
The phrase I held up was "From better production to a better society." Reading it again now, I have no intention of retracting that view. But when I wrote "production" back then, I meant, entirely, "making things." The starting point for changing society was production, and everything began there. The physical and the digital are disconnected, so connect them. Once they are connected, the floor will change. That was how simply I thought about it.
Toward the end of the same piece, there is a line I could not yet imagine how to realize, even myself.
What matters most is building systems that do not exclude humanity but rather embrace it deeply. I am convinced that our mission is to realize production systems that reflect human thought and sensibility, and that continue to evolve.
The first half overlaps with my company's guiding principles, so I understand it. But I also wrote "continue to evolve." What is it that continues to evolve, and what makes that possible? Back then, I could not yet imagine.
My thinking has only begun to change recently. There was the experience of a robot working more or less as intended while the floor as a whole still failed to run. There were equipment anomalies, and fixes for them, that only veterans could recognize. There were discussions with manufacturers responsible for maintaining social infrastructure, and site visits. Through being involved in places where real equipment runs, including the robots we provide, I have gradually come to understand something.
We may say we are producing, but what is actually at stake is whether we are producing stably. The moment the condition "stably" is added, the center of the problem moves from the ability to make something once to the ability to keep making it.
But the story did not end there. What came into view was a stranger structure. The more sophisticated a production system becomes, the more difficult, not less, it becomes to run.
The history of industry has poured astonishing obsession into automating "making." Its emblem is factory automation. Yet while the equipment makes products automatically, what keeps that equipment running is, even today, human eyes, ears and intuition. Startup, monitoring, responding to anomalies, improvement. "This machine isn't feeling well today." On the floor, it was always one particular person who could sense that.
Why does this happen? Because, to raise production capacity, we keep adding connections between equipment, software, people, organizations and supply chains. The more connections there are, the deeper the interdependence, but our ability to observe and understand it did not grow at the same pace. What a change in one part will do to the whole gradually becomes impossible to read. The very success of production is what has been growing the complexity of operations.
And so I began to wonder whether the center of scarcity has started to shift. The most value in a civilization is created by work that removes the scarcest constraint of its time. In the industrial society of the twentieth century, expanding production capacity created enormous value. Procure raw materials, turn them into high-value products with advanced processing technology, and sell them. That was what determined the hierarchy of industries.
Of course, this is not a story of one era cleanly giving way to the next. Semiconductors, aerospace, pharmaceuticals, shipbuilding: there are still fields where the ability to make something is itself scarce and remains a barrier to entry. Conversely, even in the twentieth century there were fields, like railways, power plants and telecommunications networks, where operational capability was the core of the industry. What is moving is not the era but the center of gravity of the constraint.
That center is shifting in many mature industries. Installing equipment and possessing production capacity no longer determines an industry's hierarchy on its own. It is not that production capacity has become unnecessary. It is that it is no longer enough. What determines the hierarchy is whether you can keep complex equipment running without stopping it, and keep updating it as things change. Starting with the fields where being able to make things is no longer scarce, the question changes: who will operate it? Value is beginning to move in that direction. A shrinking workforce and a shortage of skilled veterans will only accelerate this shift.
Operations, the work of keeping things running, have been spoken of in defensive words like upkeep and maintenance. I no longer see them that way, because they are not upkeep. Equipment wears down, people come and go, and materials, demand and environment all change drastically. Within all that, operations means keeping things functioning without losing sight of the purpose, constantly putting your hands on them. Not stopping and being able to change are two sides of the same capability.
I will tentatively give this a name of my own: "operational intelligence." It is not about the equipment, nor the people, nor the procedures. It refers to the cycle itself: observe reality, judge, intervene, and learn again from the results. It exists only while it is turning; when it stops, all that remains are traces.
This idea is close to "resilience engineering," which is about maintaining necessary functions even amid change and disruption. But what I see in operational intelligence is not only a heightened capacity to adapt. It is making this cycle portable across people, organizations and machines, and feeding it back into the next round of design and creation.
It is not that there was no technology for operations. Maintenance, quality control, reliability engineering, equipment management: the accumulated body of knowledge is, if anything, thick. But it has been fragmented by equipment, by company, by the person in charge. Attempts to integrate this cycle into a single whole spanning equipment and departments have not progressed far enough.
Nor is it that there are no records. Maintenance logs, standard work instructions, alarm histories, quality data, control logic: traces of operations are, if anything, left behind in great quantity. But while they record what was done, they rarely capture what was looked at, which constraints were prioritized, and why that judgment was reached. The traces remain. But the cycle never starts turning again from them. That is why operational intelligence, even today, runs only inside the head of one particular person.
As long as the cycle is closed off inside an individual, no one can read the effects of a change. If you cannot read them, failure becomes frightening. When it is frightening, people stop touching things. A system no one touches becomes a black box, and eventually no one can change it. Civilization does not grind to a halt when it can no longer make things. It halts when it can no longer change them.
This is not only about factories. Power grids, logistics, hospitals, cities, information systems: their value to society is determined not at the moment they are built but by how they are operated afterward. Civilization is thought of as the ability to build enormous artifacts. But it is also the ability to keep them running for a long time and keep updating them as the environment changes. Humanity has automated much of routine operation and monitoring. But the judgments needed outside the procedures are still left to humans.
Why does judgment keep falling to humans? Because in operations, situations constantly arise that cannot be handled with predefined states and procedures alone. They depend on context, are never put into words, and happen outside the procedures.
Conventional automation is extremely strong within the range of predefined states and procedures. But when the premises themselves change, it is fragile, because it has no cycle for learning again from what happens outside its definitions.
What AI is newly opening up is the possibility of reading these heterogeneous traces as a single situation and turning them into material for judgment. Not replacing veterans all at once. Rather, it has become possible to give the system, little by little, the cycle of learning that until now was closed off inside human heads.
Maintenance logs, work diaries, alarm histories, verbal handovers, photos taken on site. For a long time, though they were information about the same equipment, they were left in separate places, in separate formats. AI could become a means of reading across those heterogeneous traces and connecting the present situation with past experience.
The other thing that has changed is that instead of writing every response out in advance as a procedure, we can now state the purpose and the constraints, generate candidate actions suited to the situation at that moment, and compare them.
That said, having AI write plausible reasons for a judgment is not the same as accumulating operational intelligence. What was observed, what information was referenced, which constraints were considered, which action was chosen, and what happened as a result: what matters is linking these together into a single history.
What has been recorded until now is that an anomaly occurred, that a part was replaced, that the equipment was restored. But why one focused on that sign, ruled out other causes and chose that response rarely remained. If observation, judgment, action and result are linked, the experience of operations turns from mere traces into material usable for the next judgment.
Not repeating written procedures, but observing reality, choosing actions according to purpose and constraints, and updating the next judgment from the results. What lies beyond automation is autonomy in this sense.
AI is not operational intelligence itself. But it could become a means of running, on the system side as well, the cycle of observation, judgment, action and learning that was closed off in human heads.
Once we get that far, the possibility emerges of truly storing operational intelligence itself.
One of the reasons agriculture transformed civilization was that it made it possible to store food and distribute it across the seasons. Before that, to survive, humans had to procure each day's food each day. A surplus of food freed up the time and attention that had been spent on survival, and made possible the division of labor, cities and culture.
Industry today is in a similar state when it comes to operational judgment. Today's anomalies must be observed by whoever is there today and judged on the spot. Even if the equipment itself is automated, operational intelligence is still procured daily through human attention.
If we can leave records sufficient to start the cycle turning again, judgment can cross time. Just as agriculture stored food, industry will store operational capability.
But storing operational intelligence is different from storing grain. When the equipment changes, yesterday's judgment becomes outdated. Simply preserving it as a manual is not storage. It becomes an accumulation only when it is used in reality, verified by results, and reflected in the next judgment. Storing operational intelligence is not preservation; it is continual updating.
So my interest has been redefined. Building a structure that can operate over the long term. And keeping that structure in a state where it is easy to change. I do not want to eliminate human judgment. What I want to end is the structure in which everything is entrusted to one particular person's eyes, ears and intuition, and lost along with that person.
When I wrote "production systems that continue to evolve" nearly two years ago, I thought what would evolve was the equipment and the products themselves. That was not it. What has to keep evolving is the intelligence that operates them.
There are also questions I do not yet have answers to. What is missing to turn traces into seeds? What, at minimum, is needed for the cycle to start turning again inside another person, another organization, another system? The outcome of the judgment? The situation being observed at the time? The options not chosen? The constraints that were in effect? Or the history of failures and their corrections? I can list them, but I cannot yet say which are the requirements for a seed. Until this is settled, operational intelligence will remain just a phrase.
Even so, I have learned that living on with what you have made is far harder than making it. Operations are not about protecting the status quo. It is continuing to move forward amid change without losing sight of the purpose.
And operations are not on the opposite side from making. The next "ability to make" is determined by how deeply we have been able to operate what exists now. The history of running things, fixing them and changing them is what tells us what we should make next.
Creation and operations circulate, and operational intelligence is converted into the next creative power. That would mean an industry that has been consuming experience turning into one that stores experience and compounds it. Perhaps that is where an agricultural revolution in industry begins.