The Real Shortage Is Certainty

The modern economy was built on the assumption that boring would stay boring. Routes would stay open. Inputs would show up. Disputes would remain mostly abstract. Contracts would be enforced. A factory in one country could depend on a mine in another, a shipping lane in between, and an insurer pricing risk as if the world had no mood swings. When those assumptions hold, efficiency looks like virtue and redundancy looks like waste. When they fail, even briefly, the system reveals what it has always been. A physical network, wired together by trust, time, and the ability to move matter across distance. A supply chain is a trust architecture, and trust is getting priced again. Most leaders respond to volatility by doing what they have always done, only harder. They add dashboards. They add analysts. They add meetings. They add forecasts that treat a regime change like a rounding error. This produces motion. It rarely produces control. The hard truth is that the mistake is not uncertainty. The mistake is treating uncertainty like a forecasting problem. Forecasts work best when the world repeats. They work worst when the world changes the rules. A better question is simpler and more ruthless. What would have to be true for the future that is desired to occur. What is the probability those conditions hold. Which conditions are inside control. Which ones are influenceable. Which ones are externalities that will shape the field whether anyone likes it or not. That question does not soothe anyone. It does something more useful. It forces a firm to stop confusing observation with agency. This piece is about that confusion, and about the hidden cost of living inside it. It is about why correlation and probability language, even when statistically competent, can become an operating hazard when constraints return. It is about why the future will reward those who can buy options through causal action, not those who can describe risk with prettier charts. “What would have to be true for this outcome to keep repeating.”

The Age of Cheap Dependence Was Manufactured

For a long time, dependence looked safer than it historically is. Long supply chains felt normal. Specialization felt inevitable. The cost of fragility looked small, at least until the day it was not. That smoothness was not a natural condition. It was produced. It was produced by the ability to move goods across oceans with predictable cost and enforceable rules. It was produced by a world in which many countries could optimize for efficiency, because someone else was paying

much of the bill for order. It was produced by demographics that created huge labor pools in certain places while consumption expanded elsewhere. It was produced by a period when the price of time was lower than the price of redundancy, so everyone got paid for cutting buffers. When the machinery that produces smoothness weakens, the cost of dependence rises. Not as a theory. As invoices, missed commitments, and political bargains. Shipping is a good example because it is physical and hard to romanticize. Most of the modern economy assumes the ocean is a reliable conveyor belt. When it is, businesses optimize for precision. When it is not, even temporarily, businesses rediscover buffers and substitution. They rediscover that “availability” is not a line item. It is the condition that makes all other line items meaningful. The story is not that the world is collapsing. The story is that the world is charging for things that were recently subsidized or taken for granted. Time is one of those things. Trust is another. Distance is another. In that environment, governance becomes the silent cost center. Not because governance is bad, but because governance is slow when it becomes a permanent architecture for avoiding risk, rather than a discipline for acting under risk. The enterprise starts paying for time as if time were free, and then it acts surprised when competitors buy the future with speed. Time has a balance sheet. It shows up in working capital, margin leakage, and the quiet compounding of small misses that become a large miss. It shows up in the cost of rework, the cost of expedited freight, the cost of carrying inventory early because the system cannot rely on availability later. It shows up in trust with customers, which is never measured until it is gone. A firm can be brilliant at predicting the world and still be slow in the only way that matters, which is the time between signal and action. That time is rarely called delay because delay sounds like failure. It is called diligence, alignment, and risk control. Each word sounds responsible. Each step is defensible. The problem is what happens when defensible steps become the default response to every disturbance. The cost does not appear as a single incident. It appears as a drift. A drift in lead times. A drift in inventory policy. A drift in customer expectations. A drift in the internal belief that the firm can still control outcomes, when in practice it is negotiating with externalities.

Food Is Not a Commodity. It Is Legitimacy

Energy matters because it runs the physical economy. Shipping matters because it moves the physical economy. Demographics matter because they shape the labor and consumption base that keeps systems funded.

Food matters because it is legitimacy. Food is the original stability infrastructure. A society that cannot feed itself becomes a society that must bargain, and bargaining under stress is where sovereignty leaks. In stable decades, food availability can be taken for granted because trade has made calories mobile. In constrained decades, food becomes a stack. Not just crops in a field. The entire chain that makes crops and protein possible at scale. The first vulnerability is the growing season itself. Variable weather is a production variability problem. Agriculture is an output of timing, water, temperature, and soil conditions that do not take instructions. When conditions become more erratic, yields swing. Pest pressure swings. Planting windows compress. Harvest windows tighten. Irrigation demand rises when water is scarce and fails when infrastructure cannot deliver. A small change in timing can cascade through input needs, labor needs, storage capacity, and downstream pricing. Precipitation is not a background detail. It is the water budget that turns land into yield. For most major agricultural regions, that water budget is delivered by a small number of large atmospheric engines. Two mechanisms carry much of the load. Monsoonal circulation, where seasonal heating and pressure gradients pull moisture inland in pulses that are often predictable until they are not. Jet stream storm tracks, where frontal systems deliver repeated waves of rainfall across temperate zones. Many agricultural regions live on one of these. They either depend on monsoonal timing, or they depend on the cadence of jet stream driven storms. Only a limited set of places have meaningful redundancy across both mechanisms. That redundancy matters because it reduces tail risk. When a breadbasket depends on one precipitation regime, instability in that regime is not an inconvenience. It is existential to yields. Regions with two independent precipitation paths are not immune to bad years, but they are less exposed to single point weather failure. They have a second channel through which a year can be saved, or at least stabilized enough to avoid cascading shortages. That difference has strategic consequences. It changes how insurers price risk. It changes how credit behaves. It changes how much buffer a system needs. It changes the odds that a bad year becomes a political year. History adds a second layer, and it is brutal. Long term regional changes in growing conditions have repeatedly contributed to migration and upheaval. Not because weather alone “causes” collapse, but because sustained environmental stress loads a system, narrows the range of viable choices, and turns existing tensions into catalysts. People move when land no longer yields reliably. They move when water systems fail. They move when the margin disappears and does not return. Migration is not a moral event in the first instance. It is a physical one. It flows from regions that become more sensitive toward regions that are less so. It flows from single mechanism dependence toward redundancy, resilience, and surplus.

A planner who treats food as just another commodity input is planning for a world that no longer exists. Food is stability. Food is legitimacy. It is the one input politics cannot substitute. The second vulnerability is more industrial than pastoral. Modern food is manufactured, even when it grows in a field. Feeding a population depends on the ability to procure and apply the inputs that make yields possible at scale. Fertilizer is the clean example because it is not simply a product. It is an energy derivative, a mineral derivative, and a logistics derivative. Nitrogen depends on reliable energy. Potash depends on mining and concentrated supply. Phosphate depends on supply chains that are far less diversified than most executives assume. All of it depends on bulk transport, ports, rail, and the unglamorous infrastructure that becomes precious only after it breaks. If those flows are disrupted, the soil does not negotiate. Yields decline, not with a press release, but with a harvest. Protein adds another layer of constraint because protein at scale is conversion. Feed grains are converted into animal output through supply chains sensitive to price, disease, transport, and processing capacity. When feed becomes uncertain, herd sizes change. When processing becomes constrained, backlogs form and waste rises. When refrigeration and cold chain fail, the loss is immediate. Equipment is the third vulnerability, and it is easy to miss if you have not had to own a harvest window. Modern agriculture depends on machinery for planting, spraying, harvesting, transport, and irrigation. That machinery depends on a parts ecosystem. Bearings, filters, tires, electronics, hydraulic components, pumps, control units. When those do not show up, food becomes less a function of acreage and more a function of repairability and redundancy. A combine down during harvest is not a maintenance incident. It is a yield incident. It is a cash incident. It is a political incident when enough farms face it at once. The fourth vulnerability is the procurement choreography itself. Agriculture is a long loop with upfront commitments and delayed payoff. Seed, fertilizer, fuel, maintenance, labor, storage, transport. Those bets are placed months before the harvest validates them. When the broader world becomes more volatile, those bets become harder to place and more expensive to hedge. This is where the “must be true” discipline stops being a strategic exercise and becomes a survival tool. For a region to reliably feed itself at modern standards, land and water must remain reliable enough to sustain yields under weather variability. Fertilizer and crop inputs must remain available at workable cost. The industrial base must be able to build, repair, and supply the equipment and parts ecosystem. Energy must be available to run mechanized operations and

storage. Transportation and processing must exist so output does not rot in place. The cold chain must function so protein does not become waste. Some of these sit inside control. Many sit inside influence. Several sit as externalities that must be hedged rather than managed by wish. That is the point. Food is not one variable. It is a cascade. When a world becomes more constrained, cascades are what break societies.

The Only Question That Survives Regime Change

Most plans fail because they are built backward. They start with tactics, justify them with a story, measure activity, then call outcomes “unexpected” when the environment changes. A plan that survives volatility starts with conditions. It starts with an outcome, then asks what must be true for that outcome to occur. It classifies those conditions, because classification is where honesty begins. It forces the firm to name what it controls, what it can influence, and what will shape the field regardless of anyone’s preference. Then comes the part most organizations avoid because it is politically expensive. It admits that not all conditions are equally likely to hold. Some are steady, some are fragile. A plan that requires a fragile condition to behave is not a plan. It is a prayer with spreadsheets. And where I come from the end of every prayer gets an Amen. If a condition is essential and fragile, there are only a few honest moves. Change the outcome. Change the architecture so the condition is no longer essential. Buy hedges so the condition can fail without killing the enterprise. This is not pessimism. It is adult risk management. It also changes what strategy really means. Strategy becomes an option building discipline, not a prediction contest. A firm that can buy options can act under uncertainty. A firm that cannot buy options is forced to wait for clarity that will not arrive on schedule. This is where many leaders reach for data, models, and probability language. The instinct is understandable. Probability is comforting because it looks scientific. It creates the illusion that uncertainty can be tamed by more information. Often the opposite happens. More information without intervention capability creates paralysis. It turns action into re-litigation. It turns governance into theater. It turns every decision into a debate over whose forecast is “more defensible.” The cost is not abstract. It is time. It is margin. It is cash. It is the slow conversion of a firm from actor to spectator.

What gets called risk control often becomes risk multiplication, because delay creates secondary failure.

Correlation Describes. Causality Acts

Most modern organizations are dominated by correlation and probability. They have become excellent at describing what tends to happen. They are less capable of changing what happens next. Correlation thinking does three things well. It detects patterns in stable regimes. It predicts near term behavior when conditions remain similar. It scales monitoring at low cost. Correlation fails when the world changes regimes. Under regime change, yesterday’s relationships become liabilities. A model can be “accurate” by statistical standards and still be useless, because it cannot tell you what to do that will change an outcome. This is why the modern enterprise can have more dashboards than ever and still be slow in the only way that matters. It is not a data problem. It is an agency problem. Agency is the ability to shape outcomes inside constraints. That ability depends on intervention, and intervention demands a different form of reasoning. This is where causal reasoning matters. Not because it makes for better explanations, but because it makes for better action. Causality is how uncertainty becomes options. It is how a system moves from describing the world to acting on the world. The simplest boundary line is also the most clarifying. If it cannot shape an outcome, it is not an agent. That sentence cuts through much of what is being marketed as “agentic” today. Many tools are useful. Many are not agents. They are interfaces, copilots, dashboards, suggestion engines. They can speed up work, but they cannot carry authority to shape an outcome under constraints. An agent is not defined by how fluent it sounds. It is defined by whether it can take bounded action, under explicit permission, in service of a defined outcome. It is defined by whether it can close the loop, which means it can verify that its action changed the state that mattered. Once a system can act, legitimacy becomes the choke point. In volatile environments, being right is not enough. The organization must accept action without endless re-litigation. Governance must be fast without becoming arbitrary. Legitimacy does not come from explanation. Explanation is cheap. Explanation can be retrofitted. Explanation can be theater.

Legitimacy does not come from explanation. It comes from an inspectable chain of reasoning

An inspectable chain of reasoning has properties most organizations do not require from their current decision process, which is why they struggle to require it from machines. It is bounded by explicit permission and constraints. It is auditable, so evidence and rationale can be traced. It is testable, so it can be falsified by outcomes. It is revisable, so learning changes future action. This is also where Judea Pearl’s ladder stops being an academic model and becomes an operating doctrine. Association is the world of pattern and evidence. Evidence is the discipline of measuring reality and refusing to negotiate with it. Association can tell you what tends to move with what, and that is useful. It is also where most analytics and most AI stop, because it is the easiest rung to scale. Intervention is the rung where things get serious. Intervention requires context. Context is the discipline of understanding what a signal means inside a specific system, with specific constraints and goals. Intervention asks what happens if a lever is pulled. It forces the organization to name levers, constraints, second order effects, and acceptable risk. Counterfactuals are the rung where learning becomes accountable. Counterfactuals require perspective. Perspective is the discipline of reasoning about what could have happened and what should happen next, given competing objectives and alternative histories. Counterfactual thinking separates bad luck from bad decisions. It makes governance fair. It also makes governance faster, because arguments can be tested against causal claims rather than recycled as politics. Evidence, context, perspective. Those are not philosophical words. They are the prerequisites for any system that claims it can act under uncertainty without turning into a liability. This is where the automated scientist becomes a useful idea rather than a slogan. A system that can gather evidence, select interventions under permission, and learn through counterfactual tests is a system that can adapt when regimes change. It can update causal structure as constraints change. It can convert surprises into learning rather than into blame. That is foundational capability in a constrained era, because correlation will keep breaking at the edges while causality keeps offering a way to act.

The Subscription Price of Agency

Many organizations want causal options without paying for them. They want autonomy without governance. They want speed without verification. They want agents without permission. They want trust without audit. That does not work, and it fails in predictable ways.

The first failure mode is re-litigation. When action carries risk and accountability is unclear, the system defaults to procedure. Procedure is comforting because it spreads ownership thin. It makes nobody wrong in the moment. It also makes the enterprise slow. The second failure mode is theater. Dashboards update, meetings happen, and the organization calls that control. In practice, it is observation without intervention. The firm watches itself lose time, then calls it diligence. The third failure mode is the one that will create the next wave of backlash. A system is granted authority without clear permission boundaries and without a verifiable chain of reasoning. It makes a bounded decision that is not bounded in practice. The postmortem reveals that nobody can trace how the decision was made, what evidence mattered, what constraints were applied, and what would have changed the decision. Trust collapses, and autonomy gets pulled back into the safe box of read-only recommendations. The subscription price exists to prevent those failures. It starts with instrumentation that matches decisions, not curiosity. Most organizations collect what is easy. Causal options require measuring what governs outcomes, including the points where intervention occurs and the points where verification can confirm whether intervention worked. If the effect of action cannot be observed, the loop cannot be closed. If the loop cannot be closed, control cannot be claimed. It requires data integrity treated as a governance asset. In correlation systems, bad data is annoying. In causal systems that act, bad data is dangerous. When a system can shape outcomes, data becomes a lever, and therefore a target. Lineage, integrity checks, anomaly detection, and incident discipline become part of operations, not a side project. It requires a permission architecture. Permission is the architecture of authority. Authority is the architecture of action. Action is the architecture of outcomes. Without explicit permission and escalation rules, an organization will either refuse to let an agent act, or it will let an agent act until it causes a failure that forces a retreat. Neither is a strategy. It requires verification built into the mechanism. Prediction accuracy is not control. Control is verified intervention. Did the action change the state that mattered. Did it create second order effects. Did it violate constraints. A system that cannot verify is not an agent. It is a suggestion engine with better marketing. It requires audit trails that preserve legitimacy under stress. When the world gets noisy, people lose trust. When people lose trust, governance slows. When governance slows, performance collapses. An audit trail breaks that loop by making decisions inspectable. Evidence can be seen. Constraints can be seen. Rationale can be seen. What happened can be seen. What changed next can be seen. It requires continuous upkeep because regimes change. The world does not hold still long enough for a one-time model to remain authoritative. A causal system has to be maintained like

any operating asset. Assumptions expire. Constraints change. Inputs behave differently. If maintenance is treated as optional, learning rate declines, and the system starts repeating old errors with new vocabulary. This is why the era ahead will belong to architects of agency, not collectors of insight. Insight is cheap now. Agency is scarce. Agency is scarce because permission is hard, verification is hard, and legitimacy is hard. The firms that treat those as design problems will move. The firms that treat those as talking points will wait.

The Boardroom Questions That Separate Actors From Spectators

A board can spot the difference between a firm that describes risk and a firm that can act under risk, but only if it asks better questions than the ones most governance calendars allow. When the last serious disruption hit, what consumed more money. The lack of signal, or the time between signal and action. When procurement raised a constraint, did the organization have authority to act on it, or did it route the issue through committees until the market decided for it. When schedules changed, did the system verify that the chosen action improved the outcome, or did it declare success because the meeting ended. Does your enterprise have explicit permission boundaries for actions that must be taken under uncertainty, or do you rely on escalation by emotion. When something goes wrong, can you trace the chain of reasoning from evidence to action, or do you get a story that sounds plausible and ends the argument. If you replaced a human with a machine in that decision, would the organization accept the action, and if not, why should it accept the machine. A second set of questions cuts even deeper because it touches legitimacy in the most sensitive place. If a food shock hits, or an input shock hits, or a shipping shock hits, which conditions are you assuming will hold that you do not control. Which of those assumptions are fragile. Which ones are protected by redundancy, and which ones are single points of failure. If the regional precipitation regime that feeds your supply base swings hard for multiple seasons, do you have a second path, or are you depending on a single mechanism that cannot be negotiated with. What would have to be true for your current sourcing map to remain stable, and what would it cost to be wrong. The answers are uncomfortable only if you treat discomfort as a reason to avoid reality. In operations and finance, denial is the most expensive luxury.

The Counterexample That Makes the Thesis Stronger

There is a temptation to hear this argument and turn it into fatalism. That is a mistake. There are places with real structural advantages. Places with energy, water, arable land, internal transport, and some redundancy in precipitation patterns. Places that can produce key inputs, or at least substitute them, and can move food and goods without betting everything on distant chokepoints. There are firms that have already built redundancy and have learned how to act under constraint without becoming reckless. That is the counterexample, and it matters because it proves the mechanism. The future is not a uniform decline. It is a sorting. It is a sorting by constraints and by the ability to buy options. In a sorting world, correlation becomes less valuable as a primary defense, because correlation is a description of yesterday’s regime. Causality becomes more valuable, because causality is a way to act under a new regime, learn fast, and keep acting without re-litigation. A firm that sits in a favorable geography and still cannot act under uncertainty will waste its advantage. A firm that sits in a less favorable geography and can act under uncertainty will buy itself time, and time is where options compound. This is why the argument is not about fear. It is about posture. It is about deciding whether the enterprise is built to wait for the world to be kind, or built to act when the world is not.

The Credibility Tax

A prediction is useful only if it is embarrassing to get wrong. Here is one. The next major failure that freezes “agentic” programs inside serious enterprises will not be a model that predicted the wrong thing. It will be a system that acted, or was believed to have acted, without an inspectable chain of reasoning and without clear permission boundaries. The backlash will not be about intelligence. It will be about legitimacy. The phrase in the postmortem will not be “the AI made a mistake.” It will be “nobody can explain who authorized this, and nobody can prove what would have stopped it.” That is falsifiable because it is a mechanism claim. If the next backlash is driven instead by pure prediction error, with clean audit trails and clean permission boundaries in place, then this diagnosis is wrong. Now turn the prediction around, because the reversal is where advantage sits. The first firms that scale agents in a way that survives board scrutiny will do it by treating permission, verification, and auditability as the product, not as compliance. They will treat legitimacy as throughput. They will treat re-litigation as waste. They will treat loop closure as a financial asset.

They will still use correlation. They will still forecast. They will still monitor. They will simply refuse to confuse those activities with control.

The Ending Most People Avoid

The comfort of the past was not that the world was stable. It was that the bill for instability was often paid somewhere else, or paid later, or hidden inside the noise of growth. That bill is arriving faster now, and it is arriving in places that used to feel insulated. Shipping is reminding executives that distance has a cost. Energy is reminding planners that the physical economy does not run on narratives. Food is reminding governments that legitimacy begins at the dinner table. Weather variability is reminding farmers that yield is not guaranteed, and precipitation regimes do not care what a budget assumes. When those constraints tighten, the contest is not who has the best forecast. The contest is who can act with speed and legitimacy when the forecast is wrong. Correlation describes. Causality acts. Options belong to those who can act. The world will not reward better explanations. It will reward the ability to shape outcomes. References This narrative draws on Peter Zeihan’s argument that late twentieth century globalization was an engineered security and logistics regime, not a permanent baseline, and that as demographic structure, energy and input flows, and protected trade corridors tighten, dependence gets repriced first as logistics friction, then as input scarcity and volatility, then as political bargaining and legitimacy stress. That core constraint sequence comes primarily from The End of the World Is Just the Beginning, and is reinforced by The Accidental Superpower, The Absent Superpower, Disunited Nations, and Zeihan’s ongoing briefings in Zeihan on Geopolitics. The causal and agency spine of the article is informed by Judea Pearl’s ladder of causation and the distinction between association, intervention, and counterfactual reasoning in The Book of Why, Causality: Models, Reasoning, and Inference (2nd ed.), Causal Inference in Statistics: A Primer, and Pearl’s essay “The Seven Tools of Causal Inference,” which together clarify why correlation can describe regimes but cannot reliably buy control when regimes shift. The operational consequence of volatility, meaning why systems that cannot close loops drift into delay, rework, and governance theater, is supported by Deming’s management foundations in Out of the Crisis, Weick and Sutcliffe’s high reliability lens in Managing the Unexpected, and Goldratt’s constraint logic in The Goal. The institutional and legitimacy framing, meaning why order, permission, and enforcement determine what supply chains and societies can assume, is aligned with Douglass North’s institutional economics in Institutions, Institutional Change and Economic Performance and Elinor Ostrom’s collective action and commons governance in Governing the Commons. The resilience and antifragility posture, meaning why option building beats prediction when tail risk rises, is echoed in Taleb’s Antifragile. The systems view that ties these threads into feedback, delay, and second order effects is consistent with Senge’s learning organization frame in The Fifth Discipline and Meadows’s systems primer in Thinking in Systems. Finally, the article’s enterprise specific emphasis on agents as bounded outcome shaping authority, on permission as

the architecture of legitimacy, on loop closure as the economic unit of control, and on the automated scientist as the practical synthesis of evidence, context, and perspective draws on Michael Carroll’s published work and field frameworks in The One Degree Dispatch and related writings on causal options, agentic operating models, and auditable chains of reasoning.