The Factories That Will Survive Cheap Reasoning

model that assumed a depth of experience that is no longer reliably available. That mismatch is now forcing a new architecture on the enterprise, one gate at a time. That is how the plant can still be running, and still be losing. The subject of this piece is not technology. It is a repeating failure that hides behind respectable words like governance and diligence, and the question that exposes it. What would have to be true for this outcome to keep repeating.

The belief that made sense when the factory was financed

For most of modern industrial history, a factory was not only machines and process sheets. It was an inference system made of people. When a line drifted, the best operators did not need a chart to tell them. They heard it. They felt it. They knew what the system was trying to do, and they knew what it would do next if left alone. The plant did not run because data flowed. The plant ran because judgement flowed. A facility could carry complexity because experience compressed it. The longer someone stayed, the more the plant became legible. A veteran did not memorize more facts than a novice. A veteran carried better mental models. Pattern recognition did the work that analysis would have to do later. That is why the old engagement model mattered so much. The operating system assumed a stable supply of domain specific judgement. The business model assumed it too, because the assets were designed with a certain ratio of human inferencing to mechanical repeatability baked into them. When people stayed twenty five to thirty five years, the plant could carry layers of exceptions without tearing. Then the environment changed in two opposing directions. Complexity rose, and tenure fell. The system asked for more reasoning at the exact moment it possessed less of it. The result was not only a staffing problem. It was a cognition problem. Most executives still do not name it that way, and that is part of why they keep paying for the wrong fixes.

Inferencing burden, the cost nobody budgets

Inferencing burden is the reasoning load required to keep the operation inside acceptable bounds. It is the mental work of connecting signal to cause, and cause to the right intervention, under real constraints. It is the work of deciding, not the work of doing. A plant with low inferencing burden can absorb turnover. It can absorb product churn. It can absorb imperfect data. The system is legible, and the correct move is discoverable without heroics. The plant may not be world class, but it is stable in the way that makes improvement possible.

A plant with high inferencing burden is different. Context is scattered across tools and people. Exceptions arrive faster than they can be explained. Every action requires coordination because no one trusts local judgement to be safe. The work begins to feel like translation rather than production. The system still runs, but it runs on borrowed cognition, and the bill shows up as delay. When plants lose long tenure while complexity continues to rise, organizations compensate by adding gates. They add approvals, escalation paths, and cross functional handoffs that distribute inferencing upward. It looks like control. It is often the opposite. It increases the time between signal and action, and it increases the number of people required to complete a single decision. The gates are rarely irrational. They are a response to fear. Fear of being wrong. Fear of being blamed. Fear that the system is too opaque for any one person to act safely. Each gate is defensible. The combined architecture can be lethal. When inferencing capacity drops, governance expands to fill the gap.

Why more insight can make the problem worse Most industrial modernization programs treat visibility as the missing ingredient. More sensors. More dashboards. More alerts. More analytics. This belief is not childish. It is a natural response to uncertainty. When the system feels opaque, leaders reach for information. The problem appears when the true constraint is not visibility. It is interpretability and authority. A dashboard does not reduce inferencing burden if someone still has to reconcile ten systems, argue about what is true, then wait for permission to act. In those conditions, insight becomes another input to an already saturated reasoning channel. That is why a line from the investor community matters. “Insights are now cheap.” Cheap does not mean worthless. Cheap means abundant. Abundance changes the bottleneck. When insight is abundant, the scarce resource becomes the ability to convert it into action quickly and safely. That conversion requires decision rights, clear boundaries, and an operating architecture that puts context where work occurs. If a plant has high inferencing burden, adding insight without changing decision architecture can increase confusion. It adds competing truths. It creates more debates. It creates new work for the people who already spend too much time translating. A plant can become an observatory of its own decline.

The part executives avoid saying out loud

There is a statement that should make any operator pause. A plant can still operate, and still be doomed. The word is blunt because the situation is blunt. The claim is not that such a facility cannot make product. The claim is that it cannot compete, because the decision latency required to run it safely and correctly has grown beyond what the market will pay for. This is the distinction most organizations refuse to formalize. They treat every plant as improvable through effort, discipline, and tooling. That is often true inside a certain range. It becomes false when the asset base and operating design assume a level of embedded judgement that no longer exists, and cannot be rebuilt fast enough to matter. Factories are financed promises. They are promises about throughput, quality, service, and cost. Those promises are tied to assumptions about who will be there to keep the system inside bounds. When the assumptions break, the promise breaks. The factory may still run, but it runs like a firm spending down its credibility. The human cost of this truth is obvious. Plants employ communities. Closures ruin careers. Nobody serious treats that lightly. That is why leadership teams cling to optimistic interpretations, even when the operational mechanism points to a harder conclusion. Denial is often a form of mercy, until it becomes cruelty by delay. A plant can run for years while it stops competing.

The cargo cult pattern that keeps repeating The cargo cult. The story matters because it captures a recurring organizational error. People copy the visible elements of success while missing the causal mechanism. They build the runway, they wave the flags, and they wonder why the planes do not return. Industrial modernization is full of runway building. Digital boards. Control towers. AI summaries. Workflow automation. Everything can look advanced while the system remains slow, because the decision chain is still long and authority is still unclear. The cargo cult pattern is not stupidity. It is pattern matching under pressure. Leaders see what high performing enterprises display, then they reproduce the display. The failure occurs when the displayed artifact is not the cause. A dashboard is rarely the cause of speed. Speed comes from architecture. It comes from the location of decision rights. It comes from the clarity of boundaries. It comes from whether the system makes the next action legible in time to matter.

When that architecture is missing, the organization starts paying for time as if time were free.

The credible counterexample, and the boundary it reveals

There is a counterexample that must be taken seriously. Many plants have achieved major gains through better measurement and better analysis. Statistical process control, lean methods, and modern quality systems have produced improvements that are not cosmetic. In some environments, visibility was the constraint, and the disciplined use of measurement changed outcomes. That counterexample does not disprove the thesis. It defines its boundary. Those gains tend to persist when the decision chain is short and the system is legible enough for people to act. Measurement helps when it reduces uncertainty inside a decision boundary that already exists. Measurement disappoints when the boundary is missing, and the organization must build context from scratch every time it encounters an exception. The distinction is subtle but decisive. In one case, measurement clarifies action. In the other case, measurement multiplies debate. A plant drowning in inferencing burden does not need a new chart. It needs an operating system that reduces the reasoning required to act safely.

Cheap reasoning changes the definition of work

The next turn in the story is not hypothetical. Reasoning is becoming inexpensive. Systems that can summarize, compare evidence, and propose actions are now widely available. They are not perfect, and they can be wrong. Still, their presence changes the economics of decision work. For decades, middle layers in operations existed to route information. They translated between planning and production, between quality and maintenance, between engineering and the floor. Their value was often the ability to hold context and coordinate. That coordination was real work because the system could not supply context on demand. When reasoning becomes inexpensive, routing work becomes vulnerable. Not because people are unnecessary, but because the system no longer needs as many human intermediaries to move information between layers. The organization can put context closer to action. That is where the boundary moves. Work stops being defined by titles and task bundles. Work becomes defined by responsibility boundaries, which decisions someone can make, with what evidence, under what constraints, and with what accountability for the outcome. When reasoning gets cheap, outcome ownership becomes the scarce resource.

This is the moment many firms will mishandle. They will deploy reasoning tools as another layer of insight, then keep the old permission structure. They will wonder why nothing gets faster. They will say adoption is the problem. Adoption is not the problem when authority is the problem.

The operating test that separates survivors from closures

If this thesis is correct, it should be testable without slogans. A board can ask a plant leader a simple question and learn more than a stack of dashboards can show. How long is the time between signal and action when the system is under stress. Not when everything goes right. When a constraint hits. When a line drifts. When a supplier fails. When quality begins to slip. When maintenance is late and the schedule is tight. How long does it take from the first signal to the first correct intervention that actually changes the outcome. If the answer is minutes, the plant has a chance. If the answer is days, the plant is living on borrowed time, even if it is still shipping. A second question tightens the diagnosis. How many decision gates must be crossed before the person closest to the work can intervene. If an operator sees a drift but must escalate through layers because the system does not grant authority, the plant has converted speed into procedure. Procedure will win arguments in the room. It will lose in the market. A third question exposes the real cost. When the plant misses, how much of the miss is physical, and how much is time. If the root cause narrative is dominated by waiting, handoffs, meetings, approvals, rework caused by miscommunication, and decisions deferred until the next governance cycle, then the factory is not primarily constrained by machines. It is constrained by cognition and permission.

The product that falls out of the mechanism

There is a wedge that is larger than most industrial software categories. A viability model that classifies whether a plant can compete, not whether it can run. A tool that force ranks sites by the structure of their decision systems, not by a single metric.

The point is not prediction for its own sake. The point is decision compression. Corporate leaders routinely delay the hardest calls because the story is ambiguous. Ambiguity is often tolerated because it avoids blame. A viability model that makes inferencing burden visible, and ties it directly to time and money, changes that. It also changes the dignity of the conversation. Instead of blaming people for missed numbers, leaders can name the structural mismatch and decide whether to redesign or exit early. Exiting early is painful. Exiting late is usually worse, because it drains communities through years of slow decay. A model like this should not pretend to certainty it cannot earn. It should trade certainty for speed with explicit error bounds. It should be honest about what it knows, what it infers, and what it cannot know without better evidence. It should also include the counterexample class, the plants where measurement improvements are enough because the decision architecture is already sound. If this tool exists, its first buyer is not the plant manager. It is the person who is forced to decide under time pressure. Portfolio leaders, restructuring teams, operating partners, M&A diligence groups. They have to choose, and they have to defend the choice.

The prediction, stated plainly

Over the next decade, plants will separate into two visible categories. Some facilities will get faster even if their equipment looks ordinary. They will compress the time between signal and action by reducing inferencing burden. They will put context at the edge. They will grant authority inside clear boundaries. They will treat gates as rare exceptions, not permanent architecture. Other facilities will get slower even as they buy better tools. They will add visibility and still delay action. They will keep routing decisions upward because the system remains opaque. They will treat governance as the answer to blindness. They will run, and then they will stop competing. If that separation does not occur, if plants with high decision latency remain broadly competitive despite rising complexity and declining tenure, then this argument is wrong. That is the credibility tax.

The inevitable end of the old bargain

For decades, manufacturing relied on a bargain that was rarely written down. The firm would provide stable employment, and the workforce would provide stable expertise. That stability allowed the plant to carry complexity, because experience turned chaos into pattern.

That bargain is no longer reliable. Some firms will rebuild it through retention, apprenticeship, and the slow work of cultivating mastery. Many will not. Many will operate in a labor market where tenure is short and attention is scarce. In that environment, the enterprise has only two choices. It can keep building runways, mistaking visibility for control, and then act surprised when the planes do not land. Or it can redesign the operating system of the plant around clear decision boundaries, explicit permission, and reduced inferencing burden, so the factory can remain competitive without heroic tenure. The future of manufacturing will be decided by which choice becomes permanent architecture. The time between signal and action is no longer a detail. It is the business. The firms that reduce inferencing burden will buy the future with speed.

References This essay is anchored the causal claim that declining long tenure combined with rising complexity drives gate proliferation, higher decision latency, and eventual loss of competitiveness, plus the “insights are cheap” stance and the cargo cult analogy that frames artifact copying versus mechanism change. It also draws ballast from Richard Feynman’s 1974 Caltech commencement address on cargo cult science and the error of copying form without reproducing cause , John Sweller’s 1988 paper on cognitive load during problem solving and the limits of working memory when a system demands too much mental processing , Herbert A. Simon’s bounded rationality tradition as summarized by the Stanford Encyclopedia of Philosophy, which ties decision quality to limits of attention, time, and computation , and McKinsey’s published research and interviews noting that roughly 70 percent of corporate transformations fail, often through overplanning and structural pitfalls that slow execution rather than change the operating mechanism , alongside your own published work as Michael Carroll on agency as outcome shaping and the importance of decision boundaries as the unit of work in the era of inexpensive reasoning.