The Long March of Process: Efficiency and its Discontents
REVIEW ESSAY
The Origins of Efficiency
by Brian Potter
Stripe, 2025, 384 pages
Brian Potter, the author of the Construction Physics newsletter and a senior infrastructure fellow at the Institute for Progress, has written a book with an ostensibly straightforward title: The Origins of Efficiency. As Potter tells it, efficiency brought us the abundant gifts of modernity. Doing more with less is the great miracle of our industrial age, but like most such miracles—antibiotics, the power grid, fertilizer feedstock, iPhones, and so on—our habituation to efficiency’s presence has reduced it to the mundane. Efficiency, as Heidegger noticed about “being,” appears all around us and yet feels so far from our full understanding. Potter seeks to understand, “specifically,” what is happening on a farm, inside a factory, or within a company when the costs of production fall. He seeks, in other words, to understand the very nature of efficiency
In this search, Potter uncovers much more. As I’ll explain below, efficiency is not one thing but many. What at first blush appears as a guidebook for laymen on industrial process improvement evolves into a book about history and a broader investigation into knowledge, technology, and experience. Guided by Potter, the reader’s preconceived notions about the nature of history’s flow begin to blur. It becomes more difficult to separate ideas from actions, theory from praxis, idealism from materialism. Instead, Potter invites us, by way of implication, to consider that the catalytic effect of these various syntheses is what drives the movement from one epoch to the next.
Before delving into Potter’s analyses, the means by which efficiency is achieved must be understood: production process. Potter defines a production process as “a series of steps through which input materials are transformed incrementally into a finished product.” In other words, the actions that transform your raw ingredients in a cooked meal. All production processes have five elements: (1) the method of production or the cooking itself; (2) the production rate or the speed at which you cook; (3) the input and output costs or the cost of the groceries and the finished meal; (4) the buffer or how many extra ingredients you have on hand; and (5) the variability of output or the consistency of your ability to make the same meal and make it well.
From here, Potter extrapolates the criteria of efficiency: no extraneous buffer, cheaper (and fewer inputs), no wasted steps, no wasted output (use every part of the buffalo, so to speak), no variability in output, and, lastly, scale. Meeting these criteria is efficiency’s telos: a “flow process” that “continuously transforms inputs into outputs without any delays, downtime, waiting, unnecessary steps, or unneeded inputs. A steady stream of inputs goes in, and a steady stream of completed products swiftly and smoothly comes out.”
The Forces of Production
Efficiency moves forward primarily by augmenting the transformation method of materials, i.e., by introducing newer processes into production. Mechanization has been one of the primary ways we have done this. The impact of new mechanized processes, when charted over time, takes the shape of an S-curve: a ramping up of efficiency that sees an eventual plateau in which efficiency gains level out. Pan out and the S-curves begin to stack on top of each other. This stack of curves shapes the great wave over industrial modernity that has washed over the earth. To put it in poetic terms, S-curves are Prometheus’s shoulders.
S-curves aren’t guaranteed, however. Far from it. Some technologies have more “relevant axes of performance” than a single S-curve can capture. And then there’s the relationship between process method and functionality. In perhaps my favorite graphic in a book replete with elegant graphics, Potter outlines this relationship, and though the visual cannot be reproduced here, the idea conveyed is simple enough: Functionality shapes Product Design shapes Production Process. If too tightly coupled, the relationship between functionality and production process can inhibit the implementation of new production technology; tolerances for certain critical parts in a nuclear power plant are like this, for instance. The engineering standards for such components are necessarily expensive and difficult to produce. But this assumes that a successor technology waits in the wings. Not always so, Potter reminds us.
A more provocative reason can bar an S-curve from sloping upward: “Transporting a production technology (from one facility to another, or from a pilot plant to a full-scale plant) requires adapting it to a new context, which can be expensive and time consuming.” Here we have the initial stirrings of a vital theme that runs through this book: while accumulated production knowledge can (to some extent) be generalized or extrapolated from, it is not universal. The world is fundamentally too textured. Our industrial base isn’t a Lego set or a game of Minecraft. Thus, production knowledge is hard-won and iterative. And when lost, it is recoverable not from first principles but the sweat of one’s brow. Reindustrialization, if it can be done at all, can only come at great cost. After all, the original process of industrialization, wherever it occurred, was built on the painstaking absorption not just of new technology but of the complex practices and tacit knowledge cultures needed to maximize the productive value of that technology.
Consider America’s electrical build out. Utility companies and their vendors cultivated a robust “design by experience” culture whereby engineers built larger power plants premised on familiar designs. This, in turn, allowed utilities to grow their service territory. But with larger territory came greater power balancing challenges, as electricity must reconcile supply and demand at the microsecond level. In order to skill up a workforce capable of mastering this task, the electricity industry worked with higher education to create engineering departments. Imagine the decades of multigenerational learning and experience that accrued from the late nineteenth to the mid-twentieth-century—all bent toward industrial development and growth.
Now consider where the American power system is today: most of its market operators have never experienced load growth and are scrambling to adapt to increasing demand. Rather than creating a suitable investment climate for large-scale and robust power infrastructure, policymakers obsessed over fantasy climate targets divorced from engineering reality and focused the power system on short-term efficiency gains. As a result, America lacks linemen, power engineers, transformer production, and the workforce to deliver nuclear power at scale. Waves of retirements in these industries threaten to scuttle what precious process knowledge we have held onto. Thus, the United States lacks the energy infrastructure needed for economic growth.
The picture gets clearer when Potter turns toward reducing input costs. One way to reduce input costs involves redesigning the product, and there is a science dedicated to this known as value engineering, which rests on a methodology developed at General Electric after World War II to drive down manufacturing costs. A value analyst investigates each component of a product and asks “what it does, how much it costs, and whether something else might do the same job for cheaper.” A subsequent, related field is that of Design for Manufacturing and Assembly (DFMA), an approach pioneered by researchers at Amherst College in the 1960s that sought to understand “how products could be designed so their parts could be handled mechanically”; this evolved into “general assembly recommendations for both humans and machines.” What value engineering and DFMA proved is that honing the design at every step shaves millions of hours and millions of dollars off a process.
Constraints and Consequences
Functionality will always determine product design, however, which constrains options. And product design plays such a determinate role in the production process that serious tradeoffs emerge for any design decision. Some decisions could adversely impact some other part of the production process, scuttling any efficiency gains made through a given redesign. Some changes aren’t worth it, while some changes may unlock yet more changes that were previously unimagined.
Another tactic for slashing input costs involves changing the organizational structure of the production process. For instance, does a manufacturing firm need to vertically integrate every single part of the process? Maybe scaling laws would demand it, but maybe certain parts of the process can be contracted out. Henry Ford was famously a pioneer of vertical integration, but as the auto industry matured, his attachment to the practice cost him the ability to innovate later on.
But going the other way would come with its own risks as well: are there reliable contractors to be found who can actually do the job? Apple confronted this problem when it first pivoted to China. Its solution was to commit hundreds of billions in resources to training third-party vendors. Depending on the situation, inking deals with foreign contractors may harbor geopolitical risks that could create massive pain later on, something Apple didn’t consider when it moved production to Asia.
Related questions abound: could a one-off event impair production so badly the risk doesn’t pencil out? Is leaning into technological development the answer? Will the firm explore a new horizon of innovation, or exploit a tried-and-true pathway for all it’s worth? In the 1980s, GM bet that full automation was the future, whereas Toyota stuck with “autonomation,” which retained more of the human touch. In addition, the Japanese auto giant retained its kanban production method, which allowed it to get more despite using less cutting-edge technology than GM. Either way, the decisions taken here could shift entire paradigms.
As Potter reminds us, “Technological development often involves exploring the adjacent possible: the set of possibilities outside but near current possibilities.” And as the above historical examples show, identifying and seizing on “the adjacent possible” can unlock economies of scale, so long as the market can bear it. Or, perhaps, unlocking economies of scale forces a firm or founder to succeed at the above. Regardless, economies of scale “have historically been one of the most important mechanisms behind falling production costs.” But what makes for scaling? How do production costs get driven down while fattening production volume?
Spreading fixed costs, for one. “The greater the production volume, the more thinly the fixed costs are spread across each good, causing per-unit costs to fall.” Within the discussion of “economies of scale,” there is a discussion of “economies of scope,” i.e. “when manufacturers reduce unit costs by increasing the variety of goods a process produces,” or the variety of customers a manufacturer can serve. Diversifying rate structures for clientele is part of how utilities learned to get the most out of their megawatt-hours, thus enabling them to tap a broader customer pool, which in turn enabled utilities to finance larger power plants.
And then there’s geometric scaling. Making turbines, valves, and pumps bigger leads to more power out of the same nuclear reactor. But buyer beware if changing the size of the equipment necessarily changes its behavior. Statistical scaling is another factor. In the midcentury heyday of nuclear power, utilities scaled up their power generators, thus increasing the power supply, which then brought more customers into the power system. As a result, it stabilized utilities’ power output. In doing this, utilities leveraged Wright’s Law, which states that “increased production volume often results in the accumulation of cost-saving improvements as a producer or industry gains experience.”
Softer aspects play an important role. An economy of scale bespeaks larger firms and larger firms can pull in a greater range of talents, higher-quality supply networks, and lower wholesale prices. All of these coincide with network effects, “when a product or service becomes more useful or valuable as people use it.”
Innovation against Entropy
Taken together, economies of scale are “self-reinforcing” and create a virtuous cycle. But if such economies lose their financial and productive logic for whatever reason, the virtuous cycle turns vicious and sends businesses into a death spiral of increasing costs and poorer delivery. Potter invites us to consider the example of public transit: let’s say an externality (crime, for instance) inhibits people from taking BART in the Bay Area. That drains money out of the BART’s coffers, which means they would need to go to taxpayers for more funding, or increase fare prices. But the service remains worse than it was before, and so on.
There are also diseconomies of scale, in which administrative costs and bureaucratic bloat pile on until the production process becomes prohibitively expensive. After all, the firm itself grows with scale. Larger corporations mean more people which creates a costly bureaucracy of epistemic and informational hurdles. It takes more work and more money for the right and left hands to know what the other is doing. There are ways around this of course: Jensen Huang of Nvidia has greatly flattened his organization’s information flows. Admiral Hyman Rickover adopted a similar approach. But these examples are rare for a reason: each of these men exhibited exceptional leadership and management skills.
Demand effects and geometric diseconomies are cousins. The former happens when the hunger for more inputs naturally arises alongside increased production. The cost curve bites when increasing the size of something demands “proportionally more, rather than less, material.”
Panning out, industrial production emerges as an ecosystem of process knowledge, human capital, and social nodes. If all this process knowledge relies on constant doing, however, a major issue arises when it comes to loss of knowledge. There’s no “coming out of retirement” and stepping back into a championship match. There’s also no roadmap for which knowledge ought to be maintained at all times and which needs to be discarded. But the mistake to avoid is stepping out of production altogether, which means to forsake all process knowledge whatsoever.
The chapter that most deeply explores this is dedicated to removing a step from the production process. “Cutting a step out of the process is the ultimate efficiency improvement,” writes Potter. “Not only does it remove 100 percent of the inputs that step requires but it can also remove an entire scaffolding of support operations.” And there is no step more reliable than no step at all, “because an operation that doesn’t exist is an operation that can’t fail.”
From the guilds of the Middle Ages onwards, we have been refining our process knowledge. Over the course of the Enlightenment, knowledge and scientific rigor proliferated, deepening and disseminating production knowledge. The arrival of the accurate measurement tools that emerged through this process played a definitive role in honing the granularity of our process knowledge. Taylorism, with its obsessive tabulations of worker movements, gave way to greater precision; compounding innovations like the assembly line and interchangeable parts dissolved artisanal, step-laden processes in modernity’s acid bath of mechanical advancement.
On the grand scale, progress thus appears to have a firmly linear quality that does not forgive those who diverge from the long, grinding march of process. Yet it also simultaneously has nonlinear qualities worth reckoning with; this was one of Vaclav Smil’s insights in Energy and Civilization. He observed that not all progress looks like progress at first. Interchangeable parts, for instance, drove up costs “[d]ue to the imprecision of the manufacturing method and machine tools of the time,” which thus demanded more time and care in the production process.
Yet all of this is academic if one cannot rely on the process to reproduce the same results over time. Ensuring a repeatable standard of quality is crucial to economies of repetition. And if the output varies by too wide a margin, the product and its attendant process would likely be too expensive to scale. The causes behind this are destructive variation and misalignment, interrelated cost drivers that Potter illustrates with a single example:
[I]f each step in the pin-making factory takes exactly 10 seconds but has a 1 percent chance of failure, then 1 percent of the time a step won’t pass a pin to the next step and the downstream machinery will sit idle while it waits for a pin. So not only do we have waste from destroyed pins, but the process steps are also no longer aligned.
This can be managed in two ways: controlling the cause(s) and/or making the project more robust to variation. The former demands deep analysis of whether assignable causes or chance causes are driving variation, as each requires a different kind of solution. The application of statistical quality control can be followed by the identification of any reliability issues that derive from “a specific source: a machine setting, a work method, [or] an environmental factor,” which can then be stripped out.
But chance causes are multifactorial, that is, they derive from a cascade of slight shifts or changes across multiple steps of the process that culminate in output defects. Moving a production process indoors, for example, removes all sorts of chance causes by stabilizing the production environment. Meanwhile, making a process more robust might involve product redesign or tolerance adjustment to allow for greater variation where possible.
Ultimately, reducing variation demands both control and knowledge. Combining both makes for a potent system; this is why the Toyota Method (“kanban”) is the gold standard of production processes: it incorporates multiple feedback systems while increasing production knowledge and dialing in a continuous flow of product output, all while balancing supply and demand in real time. But there is no “perfect” process. Reducing variability, like everything else that goes into refining the production process, is itself a process of learning with no end insofar as our world retains its entropic character. Water is always looking for a way into your boat.
Throughout the book’s chapters, it becomes clear that each of these aspects of efficiency can and often do bleed into each other. Production processes work ecologically. As a result, great leaps forward in efficiency often come in bundles (all at once) or chains (cascades) that trigger feedback loops of improvement.
These bundles and chains trend toward continuous processes wherever and whenever possible. Continuous processes themselves represent the plateau of the S-curve, which, as Potter argues, can be mapped onto learning curves—the S-curve of efficiency slopes upward because we learn. The takeaway is that the more we repeat a process, the better we know it; the better we know it, the more we can improve it; the more we can improve it, the more efficient it becomes; the more efficient it becomes, the more we get for less.
Scale plays a load-bearing role in this dynamic. “Economies of scale are not only a major driver of efficiency improvements in and of themselves,” explains Potter, “but they are also a gating mechanism that unlocks other efficiency improvements by making it possible to amortize the large fixed costs that such improvements often require over a sufficiently large output.” Thus, repetition and scale beget efficiency, which has given us the modern world.
But efficiency is not guaranteed. Failures to improve can arise from political or regulatory obstacles, technical limitations, and market constraints; all these can debilitate a production process beyond improvement. Production burdens can also play a role. Potter devotes particular attention to housing construction, which has struggled to achieve automation despite the best efforts of men like Buckminster Fuller and Frank Lloyd Wright. More recent history rolls on like a graveyard of construction automation start-ups. Why?
In part because of Baumol’s cost disease, whereby wages rise with productivity in certain sectors, which “puts upward pressure on wages across the entire economy. If labor costs can’t easily be reduced, then labor costs drift inexorably upward. In addition, housing construction environments are outdoors and never uniform, which makes reducing variability and mechanization challenging. The cyclical nature of the housing market makes scaling up risky, too.
Potter keeps with his housing construction example to outline his vision for the shape of production to come. He calls it “flexible” production, which requires highly adaptable automation: technology that could both process data in variable environments and physically respond to such environments while maintaining efficiency. Such an achievement would be near miraculous, as Potter well knows. One of the largest challenges to this, however, is the information itself.
Simplified proving grounds could be the key to unlocking flexible production. Last year, the U.S. Army opened its first 3D printed barracks. The military’s drive toward functional reproducibility, uniformity, and reliability in output makes it the perfect staging ground for finally cracking automated housing production. With mounting experience and data from building printed barracks across different locales, commercial deployment may be hastened. As the suburban buildout of the postwar era proved, houses don’t need to look entirely unique to be attractive to young families. The suburban model, after all, took its own cues from GI housing. Leveraging AI to automate various local permitting requirements into the workflow of 3D printed housing could streamline the process and make it more adaptable. No doubt, anyone who can figure out how to safely (and aesthetically) ramp up housing in desirable locations could drop the cost of housing severalfold, ameliorating a tense political fight over housing in the process.
Efficiency as a Historical Process
As I write, the frontier AI labs are hurtling closer to realizing flexible production as a feasible method. And there’s likely a synergistic feedback loop between AI and physical production. Last year, I had the pleasure of sitting through a joint presentation from the U.S. Navy and Palantir’s Operative Systems team. They unveiled an OS for jet engine manufacturing and maintenance so impressive that, looking back on it now, makes Potter’s dream feel nearer than it is far.
Whether or not flexible production can come true is up to the future. What his book, intentionally or not, taught me about history was profound. We achieve efficiency through an accumulation of process knowledge, which we acquire through both doing and thinking. The great lesson of Potter’s book is this: humanity has not brought forth new epochs in ivory towers alone, nor have new eras been forged in the isolated confines of the factory. Rather, human history has been hammered out on an anvil made from thought and action alike.
While Potter furnishes his books with plenty of relevant evidence to persuade a reader that this is so, I found myself returning to my own bookshelves whenever I came across his examples. For instance, the culture of critique that opened up more phenomena to investigation sent me to my books detailing the Reformation, the dawn of bourgeois ideology, and the fall of the absolutist state; about Hobbes’s inspiring visit with Galileo; about sailors encountering the white waste of the Arctic, which literalized Protestantism’s hostility to adornment whilst intimating at a tabula rasa state in place of fixed nature; about Descartes’s cogito ergo sum and its strange parentage in alchemical inquiry—a parentage shared with the scientific method that enabled the precision instruments so vital to Potter’s history of efficiency.
In fact, so often did Potter send me toward other works that it slowed my progress through his own. His book had the effect of affirming my personal outlook, inclined as I am not to separate idealism and materialism but to always search for their comingling in civilization’s arc.
Yet Origins of Efficiency also gave me great pause. To hear industry boosters and some politicians tell it, a great industrial renaissance is underway in America. Many in the venture capital space have begun to pivot toward “hard tech,” taking their cue from Musk’s SpaceX. A brash “trample the dead, hurdle the weak” culture has taken root in places like the Gundo in California, where various defense, heavy industry, and energy start-ups lead the reindustrialization charge.
Much of this is inspiring, a welcome shot in the arm for otherwise droopy industries beholden to comfortable incumbents. But as we learned through Potter’s analysis, the path to efficiency always stretches beyond the horizon and entails pain and hazard. Do the summertime start-up and the sunshine industrialist have the stomach to gut out permanent enrollment in the hard school of historical-scale trial and error, participating in a process that is longer than any hype cycle or financial reporting period? I hope they do, for their sake and ours. We’ll all find out soon enough.
Too many seem to think that in a handful of years, Americans will wake up to a technofuturist version of the 1950s, when Uncle Sam’s rolled sleeves, brawny forearms, and dexterous hands peerlessly delivered the world’s goods. A necessary and uplifting vision to be sure, but let Potter pour cold water on it. After all, the United States had decades of process knowledge to build on when it tooled up for World War II, and what do we have now? What we bought dearly with experience we cashed in at the pawn show of imperial hubris. We cannot simply place a TSMC facility in Arizona and expect that it will run as it does in Taiwan because it has all the same components. People make technology, not the other way around.