Key Takeaways
Treat every optimization as a choice about what counts as best
Best is a definition. Optimization is a method for finding the best outcome within chosen limits. It begins with an objective function, the quantity to improve, then specifies available resources and constraints. A lemonade seller might maximize sweetness, profit, or customer satisfaction. Each choice produces a different recipe, and none is simply the best in every sense. The calculation can be precise while the definition of success remains debatable.
The frame shapes the answer. The book opens beside a bulldozer preparing land for Amazon's vast Kentucky air hub. The project depends on models that route packages efficiently, but those models also embody decisions about what the land is for and what costs count. Optimization is not just calculation. It is a way of seeing that can make its chosen measure look like the whole problem.
The distinction between a precise calculation and a contestable goal is central to responsible decision making. In public policy, for example, a city can optimize traffic speed while making streets less safe for pedestrians. Multiobjective methods can expose competing goals, but they cannot determine how much safety should outweigh speed. That judgment belongs to institutions and communities, not to the algorithm alone. The practical test is to ask who selected the target, which consequences are outside the model, and who has authority to revise the frame.
Measure the yield, then name what the yield leaves out
More output, fewer options. The Green Revolution, associated with agronomist Norman Borlaug, greatly increased crop production and is credited with helping prevent starvation on a vast scale. In the United States, agricultural labor productivity rose about sevenfold from 1950, compared with roughly two and a half times in other sectors. Yet farming became more dependent on costly machinery, chemicals, and uniform seeds. Soil depletion and polluted waterways were among the costs.
Bob's farm shows the bind. His North Dakota sugar beet cooperative voted to grow genetically modified beets because they promised higher yields with less labor. Bob resisted, partly to preserve his independence and familiar practices. But the cooperative controlled processing and distribution, leaving his nonmodified crop without a viable market. The book names three losses behind such gains: slack, meaning redundancy; place, meaning local knowledge; and scale, meaning a human sense of how parts affect the whole.
This account usefully distinguishes a technology's productivity from the distribution of its benefits and burdens. High yields can lower food costs and protect food security, while consolidation can transfer bargaining power away from farmers and local communities. Those outcomes are not contradictions; they occur at different scales. The argument is strongest when it resists nostalgia: small farms are not automatically sustainable, and traditional practice is not automatically fair. A sound assessment compares long term soil, income, labor, and resilience, rather than treating yield as a stand in for every good.
Break the world into parts, but keep checking the whole
Three shifts built the lens. The book traces optimization's spread through three connected habits of thought. Atomization treats a complex thing as separate units, such as bits of information or specialized factory tasks. Abstraction represents those units in models that can be compared and calculated. Automation lets machines or software act on those representations at great speed, often far from the people and places affected.
The gains are real. Claude Shannon's information theory made uncertainty in messages mathematically describable, helping lay foundations for modern computing. Factory specialization and automation also increased production dramatically. But the same methods can hide relationships among parts. A model of a power grid may capture bids and electricity flows while missing the local knowledge that reveals how a particular plant or pipeline behaves under unusual stress.
Decomposition is among the most powerful tools in science: isolating a component can reveal mechanisms that are invisible at the level of the whole. The risk arises when a useful simplification becomes an ontological claim, as if the parts were all that existed. Systems biology and ecological science respond by studying interactions, feedback, and emergence alongside components. The book's warning is therefore not an argument against models. It is a case for pairing them with observation, field expertise, and regular checks that the real system still resembles its representation.
Efficiency became a moral standard, not merely a useful tool
Thrift widened into social purpose. Benjamin Franklin treated frugality as a personal virtue: use resources carefully and organize daily life. Nineteenth century utilitarian thinkers, including John Stuart Mill, extended the question from individual conduct to society, asking how actions could produce the greatest general good. As that goal became measurable, efficiency offered a method for pursuing it. The language shifted from being prudent to making a system perform better.
The habit became ordinary. Wartime computing, industrial management, and later corporate methods made measurement and improvement part of everyday work. Marie Kondo's tidying system offers a domestic version: organize possessions through a clear procedure, with order tied to a more satisfying life. The book's concern is not that tidying or thrift is bad. It is that a method for accomplishing a chosen purpose can quietly become a moral demand to optimize every part of life.
Metrics gain power partly because they appear neutral, but institutions often attach rewards, status, and moral judgment to them. Once productivity becomes a proxy for worth, rest can look like failure even when recovery improves judgment and health. This is related to the broader problem of Goodhart's law: when a measure becomes a target, people may improve the number without improving the underlying condition. Still, quantified standards can expose waste and unequal treatment. The key distinction is whether a measure remains a revisable aid or becomes an unquestioned definition of a good life.
Keep slack in systems built to run at maximum capacity
Lean can mean fragile. Slack is spare capacity, extra time, backup equipment, or other redundancy that helps a system absorb disruption. The book's example is the February 2021 Texas freeze. Power generation, fuel supply, and household demand were tightly linked, while the state's grid had limited ability to draw help from neighboring regions. At the crisis peak, nearly five million customers lost power, and electricity prices rose from about $30 to as much as $9,000 per megawatt hour.
The design worked narrowly. Market prices responded to scarcity as intended, but that did not keep homes warm or water flowing. The book connects the blackout to a wider pattern: systems can perform brilliantly in expected conditions while failing badly outside them. A model fitted to familiar data may overlook rare but consequential events, and efficiency that removes every buffer leaves little room to recover.
Resilience engineering treats graceful failure and recovery as design goals, not as inefficiencies to eliminate. Hospitals keep reserve capacity, and forests with varied species can be less vulnerable to a single disease than uniform plantations. Such redundancy has real costs, so the choice is not simply between efficiency and waste. It is a question of which failures are tolerable, how quickly recovery must occur, and who bears the damage. The Texas example also cautions against blaming one component alone: tightly coupled infrastructure can fail through interactions among otherwise functional parts.
Do not let a measurable proxy decide a moral question
A number cannot settle fairness. Universal basic income experiments offer a way to study how cash affects people's lives. In Stockton, 125 families received $500 a month for two years, with outcomes measured. In Jackson, Mississippi, the Magnolia Mother's Trust gave selected Black mothers $1,000 a month. Its founder, Aisha Nyandoro, emphasized that targeted support addressed discrimination and a punitive welfare system, not merely a general shortage of money.
Evidence helps, but values remain. The book warns that arguments on both sides can reduce poverty policy to a question of efficient distribution. Herb Simon's term satisficing describes choosing an outcome that is good enough when perfect optimization is impractical. Yet even a well designed trial cannot decide who deserves support, what fairness requires, or which social conditions must change. Those are judgments, not outputs from a spreadsheet.
Experiments can reveal whether an intervention works and for whom, a major improvement over policy based only on ideology. But causal evidence answers a narrower question than moral legitimacy. A randomized trial can estimate the effects of a stipend; it cannot establish whether assistance should be universal, targeted, or paired with structural reform. The strongest policy process combines empirical evaluation with public reasoning about rights, history, and power. Quantification is most useful when it informs those arguments without pretending to replace them.
Reversing an optimization costs more than installing it
Systems accumulate history. The North American bison population fell from tens of millions to only a few hundred by the 1880s, as settlers killed the animals for their hides and meat and to undermine Indigenous communities. Today, hundreds of thousands live in the United States, but only a small fraction roam freely. Eastern Shoshone biologist Jason Baldes works to restore bison on tribal land, along with the cultural practices and ecosystems connected to them.
There is no simple rewind. Restoring herds means negotiating land use, water access, cattle interests, and tribal relations. The book calls this asymmetry the treachery of optimals: a system that is easy to build can be difficult to undo because people, infrastructure, and incentives adapt around it. Restoration cannot simply recover a pristine past. It requires slow, practical choices about which relationships and forms of life can be rebuilt.
Path dependence explains why early choices can constrain future options. Roads, dams, and land ownership patterns create constituencies and habits that make alternatives costly, even when the original design no longer serves everyone. Ecological restoration faces an added difficulty: species and environments change during the period of disruption, so the former state may no longer be recoverable. Baldes's work is compelling because it treats restoration as cultural and political, not just biological. The hard question is how to pursue repair without assuming that one ideal historical snapshot can dictate every present decision.
Build local refuges without pretending islands can live alone
Connection has a trade off. The book contrasts Western Samoa, a major trade stop hit hard by the 1918 influenza pandemic, with American Samoa, where officials required ships to quarantine offshore for five days. The latter's lower dependence on trade helped preserve a barrier against disease. Isolation can offer protection and room for local life, but it also limits exchange and opportunity.
The island still needs a mainland. On the author's Pacific Northwest island, baker Nathan grows wheat for his bakery, but the business still depends on ferries, railways, and sugar from distant farms. The book uses this relationship to argue for neither total integration nor total withdrawal. Local practices can preserve knowledge and variety, while broader systems provide goods and connections that small communities cannot easily supply for themselves.
The contrast resembles the resilience benefit of modular systems: separate units can contain a shock, while connections allow resources and knowledge to move between them. The balance matters. Too little connection can starve a community of support, while excessive dependence on one route or supplier creates a single point of failure. The Samoa story also shows that isolation is not inherently virtuous; it mattered because officials had information and authority to impose quarantine. Local autonomy works best when it includes the capacity to cooperate, exchange, and respond to risks beyond the local boundary.
Use optimization for bounded problems, not as a complete worldview
Neither surrender nor escape. The book rejects two tempting responses to optimization's failures: add more optimization, or try to return to an imagined earlier world. Both keep the same framework in charge. A better starting point is to recognize that mathematical tools can solve defined problems without answering every question about purpose, justice, or belonging.
Choose before calculating. The author describes leaving Silicon Valley for island life, yet acknowledges remaining connected to the systems she questioned. That unresolved position is the point. A person can use a model to improve a delivery route, then set it aside when deciding what work deserves a life. The practical move is to name the purpose of a calculation, identify what it leaves out, and choose commitments that cannot be justified by efficiency alone.
This conclusion offers a form of epistemic pluralism: different kinds of questions require different ways of knowing. Optimization is powerful for bounded tasks with explicit goals, while ethical and political choices demand deliberation among people who may value different outcomes. The challenge is institutional, not merely personal. Organizations need procedures that allow local knowledge, dissent, and revision to influence decisions after metrics have been set. A commitment beyond calculation need not reject reason. It recognizes that reason can clarify consequences without supplying the purposes those consequences should serve.
Analysis
The book's most valuable move is to distinguish optimization as a mathematical technique from optimization as a cultural worldview. The technique is not inherently ideological: a clear objective, a well specified model, and constraints can make a practical task more manageable. The worldview appears when measurable performance is treated as a complete account of value, or when the existence of a model makes its assumptions seem natural. This distinction helps the book avoid a simple anti technology argument, even as its narrative favors skepticism toward large systems. Its recurring stories make the abstract claim tangible. Bob's farm shows how a technically successful seed choice can become coercive when processing and distribution are organized around uniformity. Texas demonstrates how a market can operate according to its rules while people experience catastrophic failure. The bison restoration story reveals that systems are historically sticky: land use and cultural life cannot be reset as though they were software settings. Together, these examples show why efficiency must be evaluated across time, scale, and power, not only by immediate output.
A limitation is that the historical narrative sometimes makes optimization appear more unified than it is. Agricultural breeding, market design, and computational modeling have different aims and failure modes. Their common features do not prove that one cultural logic caused every consequence. Nor does restoring local knowledge automatically resolve trade offs around food affordability, labor, or access. These distinctions could sharpen the argument.
The book's constructive proposal is less a new method than a discipline of attention: keep the tool, resist its claim to completeness. That is compatible with resilience engineering, which deliberately designs for disruption, and with democratic governance, which makes conflicting values visible rather than hiding them inside a metric. The central lesson is not to stop calculating. It is to keep deciding what deserves calculation, who gets to decide, and what kinds of life should remain outside the optimization.
Review Summary
Optimal Illusions by Coco Krumme explores the downsides of optimization in various aspects of life. While some readers found the book thought-provoking and well-researched, others criticized its lack of cohesion and clarity. Many appreciated the interesting anecdotes and examples but felt the overall argument was not fully developed. Some readers praised Krumme's writing style and insights, while others found the book repetitive and unfocused. The mixed reviews suggest that while the topic is intriguing, the execution may have fallen short for some readers.
People Also Read
Glossary
Objective function
The goal being optimizedAn objective function specifies what an optimization is trying to improve, such as profit, speed, or output. Because different goals produce different answers, defining the function is a value laden choice as well as a mathematical step. The book uses a lemonade stand to show how maximizing sweetness can yield a different result from maximizing sales or minimizing cost.
Atomization, abstraction, and automation
Three habits behind optimizationThe book uses these terms for three shifts that helped optimization spread. Atomization breaks systems into units; abstraction represents those units through models and symbols; automation lets machines or software act on those representations at scale. Together, they enable powerful calculation and coordination, but can separate decision makers from the material settings and people affected by a system.
Slack, place, and scale
What efficiency can erodeThis trio names losses that can accompany tightly optimized systems. Slack is spare capacity or redundancy that cushions shocks. Place is knowledge and practice rooted in a particular landscape or community. Scale is an understanding of how an individual activity connects to a larger whole. The terms describe qualities that productivity measures may fail to capture.
Jevons's paradox
Efficiency can increase demandJevons's paradox describes the possibility that making a resource cheaper or more efficient to use increases total demand for it, rather than reducing consumption. The book traces the idea to William Stanley Jevons's study of coal. When mining became more efficient, lower costs encouraged new uses for coal and greater dependence on it.
Satisficing
Choosing what is good enoughHerbert Simon used satisficing to describe decision making under uncertainty when finding the global best option is too difficult or costly. Rather than search indefinitely, a decision maker selects an outcome that meets an acceptable threshold. The book presents it as a practical alternative to demanding perfect optimization, while warning that it does not settle which outcomes or thresholds are fair.
The treachery of optimals
Hard to reverse optimized systemsThis phrase describes how an optimization can be mistaken for the reality it represents, and how the system built around it becomes difficult to undo. Infrastructure, incentives, habits, and knowledge adapt to the chosen arrangement. Reversing it is not a simple return to an earlier state, especially in ecosystems and communities shaped by long histories.
Monte Carlo method
Estimate outcomes through samplingThe Monte Carlo method uses repeated random sampling to estimate the range or likelihood of outcomes in a complicated problem. Stan Ulam conceived the approach while thinking about solitaire, then saw its potential for physics calculations involving uncertain events. The method made it practical for computers to simulate many possible interactions where exact calculation was difficult.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.