Part Two: What The Boxes Were Carrying

Part two of a three-part series on the changing architecture of coordination.

In Part One, I described a small-business credit decision in which the numbers showed two quarters of weakening receivables. The person who knew the business also knew that its largest customer had temporarily extended payment terms during an acquisition while orders remained strong. The numbers were accurate, but they did not carry the full meaning of what was happening.

The relationship did. It connected the current numbers to the history of the business and gave someone a basis for recognizing that apparent deterioration meant something different in this case.

For much of the Industrial Age, organizations accumulated enormous amounts of this kind of coordination capacity without naming it that way. It lived inside roles, departments, professional communities, routines, records, reporting relationships, and repeated interactions among people who learned the system and one another over time. As work becomes more automated and more distributed, some of that capacity can disappear with the work that once carried it.

The coordination the organization chart missed

Hierarchy and organizational boundaries provided an important part of industrial coordination. They established where work belonged, who had authority, where responsibility sat, and how an issue could be escalated when the normal process stopped working. But much of the coordination inside an organization was less formal.

A role created repeated exposure to the same domain. People learned which signals mattered, which exceptions were harmless, which were dangerous, and when an apparently routine case deserved another look. Departments concentrated expertise, but they also created communities where people compared experience and absorbed ways of thinking that were difficult to encode in a procedure.

Processes contributed as well. A review step might have existed to check compliance with a rule, but it also created a predictable place for another perspective to enter. Records preserved what happened, while relationships often preserved why it happened.

Over time, these mechanisms allowed organizations to carry knowledge that was only partly written down. They created continuity between one decision and the next and gave people a larger frame for interpreting new situations. None of this appeared clearly on the organization chart, yet the organization depended on it.

Automation can remove more than the work

The case for automation usually begins with the visible activity. If software can perform a recurring step faster, more consistently, or at lower cost, that step becomes a natural candidate for removal. The difficulty is that the visible activity and the coordination surrounding it are not always the same thing.

A person reviewing an application may have been checking whether required information was present, but that review also gave someone a chance to notice that the information did not fit together. A recurring customer conversation may have been expensive to deliver, but repeated conversations were also how history and trust accumulated.

When the task disappears, the productivity gain is immediate. The lost coordination capacity is harder to see because it was rarely measured in the first place.

This becomes more consequential as work is divided among digital channels, models, automated agents, external providers, and specialized systems. Each component can perform its assigned function extremely well while knowing very little about what the other components know. The system becomes stronger at executing the parts while becoming more dependent on how well those parts are connected. Context is one of the first places that dependence becomes visible.

When context stops accumulating on its own

Context is more than additional information. It connects information to the circumstances that give it meaning.

The small-business numbers tell us that receivables weakened. Context tells us why. An engineer can see that a reading falls within an acceptable range and still recognize a pattern that resembles the early stage of a failure encountered years earlier. In both cases, the value comes from relating the immediate signal to something outside the immediate event.

Industrial organizations often accumulated this kind of understanding almost incidentally. People stayed in roles long enough to see patterns repeat. Customers returned to the same institution. Teams worked together across many cases. History attached itself to relationships.

A more automated system cannot assume that context will continue to accumulate in the same way. If customer interaction becomes thinner, work moves among systems, and decisions are assembled from data held in different places, the organization has to decide deliberately which context travels with the work. That immediately raises questions of legitimacy as well as capability. Which information is relevant? Where did it come from? How reliable is it? Who is allowed to use it? Can the person affected see it, correct it, or challenge the interpretation?

The simplest technological response to context loss is to collect more information. That can improve understanding, but it can also create a system that knows more about people than it has a legitimate reason to know. Better context and greater surveillance can emerge from the same technical capability.

Stable human relationships once handled some of those boundaries informally. A person might know a great deal about a customer while also understanding that much of it had no place in a particular financial decision. Once context becomes data, those boundaries have to become explicit.

The apprenticeship problem is also an economic problem

Another hidden function of industrial work was development. Organizations did not only use experienced people. They produced them.

A junior employee handled ordinary cases, watched more experienced colleagues, made small mistakes, received feedback, and gradually encountered more difficult situations. Much of that learning happened because the work had to be done anyway. The organization received productive output while the employee accumulated experience.

That made apprenticeship unusually cheap. Learning was embedded inside economically necessary work, so it did not always have to be separately funded, scheduled, or defended as an investment.

Automation changes that arrangement. If AI absorbs a large share of routine work, the cases left to humans become more ambiguous and more consequential at the same time that the lower-risk experiences through which people historically learned to handle them begin to disappear.

Higher-value work usually requires capabilities developed through lower-risk experience. Once routine work has been removed, creating those experiences becomes a separate cost. A manager has to allocate time to it. Someone has to justify the investment. The immediate economics encourage the organization to capture the productivity gain while postponing the development investment that the new model requires.

One promising response is to make disagreement between people and machines part of the new apprenticeship. A developing professional can make an assessment, compare it with the system’s conclusion, and then examine the difference with someone who has deeper experience. The learning sits in the gap. What did the model notice that the person missed? What context changed the expert’s interpretation? Which assumption drove the result? When was the machine technically correct but incomplete?

That creates repeated exposure to judgment without requiring people to spend years performing every routine task manually. The machine becomes part of the learning environment rather than simply the mechanism that removed the old one.

Organizations that automate the training ground will have to invest deliberately in another way of producing expertise. Otherwise, they risk becoming highly efficient consumers of judgment while steadily weakening their ability to create it.

Some friction was doing coordination work

Organizations have spent decades removing friction, often with good reason. Customers should not wait unnecessarily. Employees should not chase approvals that add no value. Information should not sit inside inaccessible silos. But some friction was serving another purpose.

A second review can be needless bureaucracy, or it can be the moment when a flawed assumption is challenged. Two colleagues comparing partial observations can look like an inefficient interruption, or it can be where separate weak signals become a pattern neither person could see alone. Those situations may look similar on a process map, but they are performing very different functions.

If a step merely consumes time, remove it. If it was carrying context, judgment, trust, learning, or the ability to interrupt failure, removing the step transfers that obligation somewhere else.

The question is whether the new operating model knows what obligation it has inherited.

Separate the container from the capacity

None of this requires preserving the industrial organization in its current form. Many of its structures accumulated long after their usefulness declined. Boundaries trapped expertise. Hierarchies slowed decisions. Processes became rigid. Customers and employees absorbed the burden of navigating structures designed around institutional convenience.

AI and other digital capabilities create a real opportunity to remove much of that burden. The harder work is separating the familiar container from the capacity it provided.

A relationship manager is one way to preserve context, but context can be preserved in other ways. Managerial approval is one way to establish authority, but authority does not have to depend on moving every decision upward. Routine work was one way to develop judgment, but future expertise does not have to depend on people spending years performing tasks machines already handle well.

Once the container and the capacity are separated, the design question changes. The issue is no longer how much of the old structure to preserve. It is which capacities still matter when the structure changes, and how deliberately the new system provides them.

Context, accountability, and judgment do not have to remain attached to the structures that once carried them. But they do have to exist somewhere.

Part Three will look at what that new coordination system has to provide when people, machines, expertise, information, and authority no longer sit neatly inside the same organizational boundaries.