In 1986, industrial technology looked very different. Factories were becoming more automated, but they were still dominated by specialized systems built to control specific machines and processes. Distributed control systems, programmable logic controllers, drives, instrumentation, and plant-floor computing were changing what manufacturers could see and manage, but many of those technologies still lived in relatively isolated worlds. The plant floor had its systems, the enterprise had others, and warehouses, transportation networks, suppliers, and customers operated through still more layers of technology, many of them disconnected.
It was into that world that Automation Research Corporation, later ARC Advisory Group, was founded. At the time, one of the central industrial technology questions was how far computing could reach into the physical systems that actually ran factories, utilities, energy infrastructure, and industrial operations. Almost forty years later, we know the answer: it reached nearly everywhere.
Now a new question is taking its place. What happens when intelligence does the same?
For Logistics Viewpoints, as part of ARC Advisory Group, that is not simply another AI story. It is the next chapter in a much longer technological progression that ARC has been watching since its beginning. Industrial technology learned first to control physical processes, then to connect them, then to see what was happening across them. Eventually, it learned to predict what might happen next.
Now it is beginning to decide what should happen next—and, increasingly, to act.
The Machines Learned to Talk
The first great transformation was the automation of physical processes. Machines became increasingly instrumented, control systems became more sophisticated, and industrial companies gained finer control over production, quality, throughput, and equipment performance. But automation had limits. A system might know exactly what was happening inside one production process while remaining almost completely blind to what was happening elsewhere in the organization.
The next breakthrough, therefore, was connectivity. Over time, systems that had once functioned independently began to communicate. Plant-floor systems connected with manufacturing applications, manufacturing information flowed into enterprise software, industrial Ethernet spread, standards improved, proprietary environments became more open, and operational technology and information technology began slowly moving toward one another.
That transition took years, and it rarely unfolded as neatly as technology presentations suggested it would. Factories could not simply stop running while their technology stacks were replaced. Refineries could not be rebuilt every time a new software architecture appeared. Warehouses, utilities, railroads, and global supply chains carried decades of accumulated infrastructure with them, so new technologies had to work alongside old ones.
That became one of the enduring lessons of industrial technology: invention matters, but integration determines whether the invention becomes useful.
Then Everything Began Producing Data
Once the industrial world became more connected, another transformation followed almost automatically. Connected systems produced information—enormous amounts of it. Machines, production systems, orders, warehouses, trucks, suppliers, and customers all became sources of data. At first, simply collecting and displaying that information created tremendous value.
Supply chain technology illustrates this particularly well. For years, companies struggled to understand what was happening beyond the walls of their own facilities. A shipment could leave a supplier and effectively disappear into a transportation network until it arrived—or failed to arrive. Transportation visibility platforms began changing that. Warehouse management systems provided increasingly detailed operational insight, control towers attempted to bring information from multiple systems into common views, supplier platforms digitized procurement relationships, and planning systems connected demand, inventory, production, and replenishment.
The industry spent enormous amounts of money trying to see the supply chain more clearly. And it worked, perhaps too well.
Visibility Created a New Problem
A modern supply chain can now tell you an extraordinary amount about itself. A shipment is running late. A supplier has missed a commitment. Inventory is below plan. A vessel is delayed. Demand is accelerating. A carrier has rejected a load. A distribution center is falling behind. A customer order is at risk.
Every one of those events can generate an alert, and a large enterprise can generate thousands of them. That creates an uncomfortable realization: the problem is no longer simply that organizations cannot see what is happening. Increasingly, the problem is that they can see too much.
Somewhere along the way, visibility ceased to be the final objective and became the beginning of another problem. Once you know that something has gone wrong, someone still has to determine whether it matters, understand why it happened, evaluate the alternatives, make a decision, and then act. For decades, that “someone” was usually a person.
Now, for the first time, that assumption is beginning to change.
AI Enters the Operating Model
Most of the public discussion around artificial intelligence has focused on what AI can generate: text, images, code, answers, and summaries. Those capabilities are significant, but in industrial settings they may turn out to be the least important part of the story. The larger change begins when AI starts participating in operations.
Imagine a delayed inbound shipment. A traditional visibility system tells the planner that the container will arrive three days late. That is useful information, but the real business questions begin immediately afterward. What is inside the container? Which plants need those materials? Which production orders depend on them? Which customer commitments could be affected? Is replacement inventory available somewhere else? Can another supplier provide the component? Could production be resequenced? Would expediting another shipment cost less than interrupting production?
Ultimately, the organization needs to know which response produces the lowest total business impact while protecting its most important commitments. That is no longer simply a visibility problem. It is a reasoning problem, and increasingly it is becoming an AI problem.
From Information to Understanding
This is where the emerging architecture of industrial AI begins to matter.
AI needs more than a powerful model. It needs access to the information, relationships, tools, and operating context of the enterprise. Retrieval-augmented generation can ground AI in company-specific knowledge. Graph-based approaches can help it understand relationships among suppliers, products, plants, shipments, customers, and orders. Agent-to-agent communication can allow specialized AI systems to coordinate, while emerging protocols can connect those systems with enterprise tools and contextual resources.
ARC’s work on AI in the supply chain describes this as the foundation for a connected intelligence layer spanning existing enterprise and operational systems. The important point is not the collection of acronyms surrounding AI. It is what the architecture makes possible: moving from an AI system that answers an isolated question toward one that understands what an event means within the broader operation.
That distinction is particularly important in supply chains because almost nothing exists independently. A supplier is connected to components. Components are connected to products. Products are connected to plants. Plants are connected to customers. Shipments are connected to orders, and orders are connected to revenue.
A late container is therefore not simply a late container. It may be the beginning of a chain reaction.
And once a system can understand that chain reaction, the next question becomes unavoidable: should it act?
Software Begins to Cross the Line
For most of the enterprise software era, the division of labor between humans and systems was clear. Software stored transactions, calculated plans, displayed exceptions, produced forecasts, and recommended options. People connected the dots. A planner saw the problem, investigated the cause, called another department, evaluated alternatives, made a decision, entered the change, and watched what happened next.
Agentic AI begins to blur that division. Consider the same delayed shipment. A shipment agent detects the delay while an inventory agent determines which locations will be affected. A production agent identifies manufacturing orders at risk, a procurement agent searches for alternative supply, and a transportation agent evaluates expedited options. Those systems can potentially exchange information, compare alternatives, and weigh cost, service, inventory, and production consequences.
If the problem falls within predefined authority limits, the system could eventually take the action itself. If it does not, it could send the issue to a human with much of the investigative and analytical work already completed.
That is not simply better analytics. It represents a different operating model because software is beginning to cross the boundary between supporting work and performing work.
Forty Years in One Line
Seen from the perspective of 1986, the progression is striking.
Automation → Connectivity → Visibility → Prediction → Decision → Action.
Automation gave industrial systems the ability to control physical processes. Connectivity allowed those systems to exchange information. Visibility gave organizations a clearer understanding of what was happening across increasingly complex operations. Predictive analytics helped anticipate what might happen next. Decision intelligence began recommending what organizations should do about it.
Now agentic AI is beginning to explore the final step: taking action.
It is not a perfect description of every technology category, and not every operational process will move through all six stages. Nor should every business decision become autonomous. But the framework captures the larger direction of travel.
What makes the current moment especially interesting is that the final two stages are developing together. Decision intelligence and agentic AI are advancing at the same time, meaning the gap between knowing what should happen and actually making it happen could begin to shrink dramatically.
And Then Reality Arrives
Of course, every technology revolution looks easiest before it touches an operating environment. An AI agent resolving a simulated supply chain problem is impressive. Giving that agent authority over a real production schedule, inventory allocation, transportation move, procurement decision, or customer commitment is something else entirely.
Suddenly the questions become less glamorous and much more important. What information can the agent access? Which systems can it modify? How much money can it commit? Which decisions require human approval? What happens when two agents optimize for different objectives? How do you reconstruct a decision six months later? Who is responsible when the system is wrong? And how does this sophisticated new intelligence layer interact with an ERP implementation that may be fifteen years old or a warehouse system that the business cannot afford to replace?
These questions may sound like constraints, but they are actually where industrial transformation happens. ARC has seen variations of this problem repeatedly. Open systems did not make proprietary infrastructure disappear overnight. Cloud computing did not eliminate enterprise systems. Industrial IoT did not replace control systems. Digital transformation did not sweep away decades of installed technology. Instead, every new layer had to find its place within what already existed.
AI will be no different.
The Future Will Be Built on the Past
There is a tendency in technology to describe every new era as though everything before it suddenly became obsolete. Industrial technology rarely works that way. The future tends to accumulate rather than replace: new layers sit on top of old ones, new architectures connect with installed systems, and new intelligence depends on existing transaction platforms.
AI will not erase ERP, TMS, WMS, procurement, planning, or control systems. It will increasingly operate across them. That is why the AI conversation inside industry will eventually become less about models and more about architecture: data harmonization, interoperability, security, governance, context, decision rights, and human oversight.
Those are the mechanisms through which AI stops being an impressive demonstration and becomes part of the operating fabric of the enterprise. The real competitive advantage may ultimately come not from having access to the most powerful model, but from connecting intelligence successfully to the data, systems, workflows, and decisions through which the business actually operates.
The Categories Are Starting to Move
We can already see the consequences in the supply chain software market. Visibility platforms are moving toward exception management. Exception management is moving toward decision intelligence. Decision intelligence is moving toward execution. Planning systems are incorporating generative AI, transportation and warehouse platforms are beginning to add autonomous capabilities, and control towers are evolving toward orchestration environments.
Software categories that once seemed distinct are beginning to overlap. That makes this a particularly important moment for technology research because mature and emerging markets require very different kinds of analysis. In an established category, the questions are familiar: Who are the leaders? How large is the market? What features does each supplier offer? How quickly is the category growing?
Emerging markets create harder questions. What exactly is the category? Where does it begin and end? Which capabilities actually belong inside it? What architecture is required? Which decisions should remain advisory and which can become autonomous? What level of buyer maturity is necessary?
Before suppliers can be compared, the market itself often has to be defined.
That has always been one of the most important functions of industrial technology research, and it becomes particularly important when the underlying architecture is changing as quickly as it is today.
Back to 1986
And so, after nearly forty years, the story comes almost full circle.
When ARC began, the defining question was what would happen as computing became deeply embedded in industrial operations. The answer unfolded over decades as machines became automated, systems became connected, operations became visible, data became pervasive, and analytics became predictive.
Now the industry is moving into another phase. The question is no longer simply whether industrial systems can collect information or even understand what is happening. The question is whether they can increasingly determine what should happen next and whether, under the right circumstances, they should be allowed to make it happen.
For Logistics Viewpoints, that makes the AI era particularly significant. We are not watching a technology category appear in isolation. We are watching the next stage in the evolution of the systems that run supply chains and, more broadly, the physical economy.
ARC has spent nearly forty years studying that evolution. The technologies have changed, but the underlying question has not: How does a breakthrough in computing become something industry can actually trust, integrate, and use?
In 1986, that question began with automation. Today, it begins with intelligence.
The next industrial era will be defined not simply by systems that can see more or predict more, but by systems increasingly capable of deciding and acting. For ARC, that makes AI less a break with its past than the logical continuation of it: another fundamental change in the architecture of the physical economy, and another transition whose real significance will only become clear when the technology meets operations.
The post From Automation to Intelligence: ARC’s Next Industrial Technology Chapter appeared first on Logistics Viewpoints.