Key Takeaways
Measure efficiency by durable capability, not by immediate speed
Speed hides costs. Efficiency usually means getting a task done with less waste. Tenner’s paradox is that optimizing for speed now can create waste later: users lose skills, organizations miss opportunities, and systems become brittle. He calls the alternative inspired inefficiency: deliberately preserving activities that may look slower but build lasting capability, such as learning to navigate without relying entirely on GPS.
Seven failure paths. The book identifies suppressed serendipity, narrowed focus, cascades from early decisions, eroded human skill, gaming of performance measures, overwhelming data, and systems standardized beyond their ability to adapt. These risks appear across search, education, transport, and medicine. The point is not to reject technology, but to count the costs that a narrow measure of speed or output leaves out.
What's striking is that this reframes efficiency as a property of a whole system over time, not a single task performed quickly. That connects to resilience engineering, which asks whether a system can absorb failure and adapt, rather than merely run smoothly under ideal conditions. The distinction also helps explain why a slower backup skill can be valuable even when rarely used. The framework has a limit: preserving every old practice would be wasteful too. The useful question is which forms of friction create learning, flexibility, or recovery capacity.
Reward market-making breakthroughs, not only frictionless platforms
Matching is not invention. Nineteenth-century continuous production transformed what people could buy: Oliver Evans integrated milling operations into a continuous process, while new papermaking machinery made paper far more abundant. Many modern platforms do something different. They use software to match existing buyers and sellers more cheaply, as rideshare services match passengers with drivers. That can be convenient, but it does not necessarily create a new capability or reduce the need for travel.
Innovation needs slack. Tenner argues that money and talent can flow toward platform businesses because their results are easier to test and scale than difficult physical innovations. Yet breakthroughs such as the photocopier required years of costly experiments and setbacks. The Hiding Hand describes how project leaders may begin without seeing every obstacle, then discover ways to overcome them. That uncertainty is risky, but eliminating it can also eliminate important inventions.
The distinction between making transactions cheaper and creating new markets is a useful corrective to treating every successful app as a broad social advance. It also points to a measurement problem: company growth is easy to observe, while the opportunity cost of neglected research is harder to see. Still, platforms can create real value by widening access and reducing search costs. The sharper test is whether they merely rearrange an existing activity or enable people to do something they could not practically do before.
Give algorithms room to surface the unlikely, not just the already popular
Rankings compound advantage. Google’s PageRank made search more useful by evaluating how web pages linked to one another, adapting an approach developed to track scientific citations. But links and recommendations can turn small early advantages into huge ones. The Matthew effect names this cumulative advantage: sources that gain attention are more likely to receive still more attention. Tenner describes the same pattern in science, where researchers increasingly cite popular recent work, and in music, where recommendations can magnify an early lead.
Popularity is not proof. Research on an online music market found that popularity and listeners’ independent quality judgments were related, but only modestly. The book’s practical implication is to look beyond the top-ranked results and make room for less familiar sources. A slower, more varied search can reveal useful work that a popularity-driven ranking leaves buried.
Recommendation systems solve a genuine coordination problem: without ranking, abundance can make discovery exhausting. The danger is that a ranking becomes part of the world it claims only to describe. Once people follow the visible leaders, popularity supplies its own evidence of worth. This is a form of path dependence, where early conditions shape later outcomes. Diversity controls can help, but novelty alone is not quality. A strong discovery system should expose alternatives while preserving ways to assess their reliability.
Audit the proxy before it starts replacing the purpose
Measures change behavior. Campbell’s Law describes how a measure becomes less trustworthy when people use it to make consequential decisions, because they have incentives to manipulate the measure. A school that prizes test scores can end up teaching test performance instead of understanding. Online reviews invite fake praise, and medical billing codes can reward providers for classifying cases as more complex than they are.
Keep the goal visible. Tenner argues that data and metrics can still improve decisions, but they should be treated as partial evidence, not as the goal itself. In science, for example, citation counts can help identify influential papers, yet they also reward already popular research and may encourage attention-grabbing findings over careful replication. A useful check is to ask what the measure leaves out, who benefits when it rises, and whether the measured behavior still serves the underlying purpose.
This is closely related to Goodhart-style effects: when a measure becomes a target, people adapt to the target, and its relationship to the original goal can weaken. The insight matters because dashboards often make the proxy look objective while hiding the choices behind it. Measures are not useless, however. They can reveal patterns that unaided judgment misses. The safer approach is to combine them with independent checks, qualitative evidence, and periodic review of whether the indicator still tracks what matters.
Make learning difficult enough to require thinking
Effort can aid memory. Desirable difficulty is the idea that some manageable effort improves learning by making the learner process information more deeply. In experiments described by Tenner, students taking notes by hand remembered concepts better than students typing more complete notes, even when online distractions were removed. Handwriting forced them to select and restate ideas instead of transcribing every word.
Use media for different jobs. The book presents screens as useful for searching and retrieving details, while print can help readers grasp relationships and larger arguments. This does not mean that paper is always superior or that frustration is educational. Teachers, rather than devices alone, remain central: effective online lessons can take many hours of skilled preparation, and schools with fewer resources may not have the support needed to use technology well.
The distinction between productive effort and pointless friction is crucial. Learning research often finds that retrieval, paraphrasing, and spaced practice strengthen later recall, but a task that is merely confusing can consume effort without building understanding. The best design does not make everything hard. It removes mechanical obstacles while preserving the thinking the learner needs to do. That is why a tool’s value should be judged by what students can explain or apply afterward, not by how modern or convenient the lesson feels.
Use navigation apps without surrendering your mental map
Directions are not orientation. GPS can locate a traveler precisely and offer a fast route, but following turn-by-turn instructions does not necessarily build a mental picture of the surrounding area. Studies discussed by Tenner found that people using directional prompts remembered their routes less well than people using paper maps. Inuit hunters combine GPS with knowledge of ice, weather, and terrain, sometimes rejecting the device’s direct route when local conditions make it dangerous.
Keep a backup skill. London taxi drivers traditionally spent years learning the city’s streets, building detailed knowledge through repeated travel. That expertise is not necessary for every driver, but the example shows that spatial awareness can be trained. Checking a route before departure, noticing landmarks, and occasionally navigating without prompts can help preserve useful orientation skills, especially when a device fails or its instructions conflict with the visible surroundings.
Navigation illustrates a wider principle of human-computer interaction: tools can reduce immediate cognitive load while also reducing practice. GPS is often an excellent choice, particularly in unfamiliar or urgent situations. The risk is not using it, but treating its output as a complete account of reality. A route can be correct in the database and unsafe in context. A sensible design gives users both a recommended path and enough context to notice when conditions on the ground call for a different decision.
Keep medical decisions human, even when the data are excellent
More records, more work. Electronic medical records were meant to improve communication and reduce errors, including those caused by illegible prescriptions. Yet one study cited in the book found physicians spending about two hours on administrative work for every hour of direct patient contact, with additional work often continuing at home. Medical systems also generate so many alerts that staff can become desensitized. At one hospital, intensive-care alarms averaged roughly one every eight minutes per patient.
Data need interpretation. Tests can reveal useful risks, but they can also produce false positives, incidental findings, and anxiety that leads to unnecessary treatment. Algorithms may help detect rare conditions, yet they cannot fully replace a clinician’s understanding of a patient’s circumstances, priorities, and response to care. Tenner’s case is not for returning to paper records or relying on instinct alone. It is for combining data with time, listening, and professional judgment.
Medicine makes the costs of narrow efficiency unusually visible because the patient is not a passive object being processed. More information can improve diagnosis, but it also creates decisions, follow-up tests, and emotional consequences. The medical setting therefore calls for shared judgment: clinicians contribute expertise, while patients contribute knowledge of their own lives and tolerance for risk. Technology can support that exchange, but systems that consume the time needed for it may undermine the very care they are designed to improve.
Automate at the midpoint, where people can still recover
Keep a capable operator. The Perfect 5 is a model that places automation midway between fully human and fully automatic control. The idea is not that every system should be exactly half automated, but that people should remain able to understand and intervene. In aviation, highly reliable autopilots reduce routine workload, yet rare failures can leave crews needing skills they have had fewer chances to practice. Air France 447 crashed after instrument problems shifted control back to pilots who faced confusing signals and failed to recover the aircraft.
Practice the handoff. Automated cars and medical systems raise similar concerns: an operator who normally supervises a system may not be prepared to take control under pressure. Systems should make their state legible, provide meaningful opportunities for intervention, and train people for failure conditions. Automation can improve consistency and safety, but a system that works only when nothing goes wrong is not robust in the long run.
The midpoint idea aligns with resilience engineering, which focuses on how people and systems respond to surprises rather than only preventing routine errors. Human oversight is valuable only when it is informed and practiced. A person asked to monitor an uneventful process for hours may be poorly positioned to detect a rare failure. Thus, keeping a human in the loop is not enough by itself. The loop needs clear signals, authority to act, and rehearsal of the moments when the machine needs help.
Build in six kinds of inefficiency that improve the whole system
Choose useful friction. Tenner’s conclusion offers six ways to make technology more capable rather than merely faster. They can be grouped into five practices:
1. Keep automation near a human-controlled midpoint, with trained operators.
2. Preserve physical presence and tangible representations when they improve context.
3. Allow creative waste: experimentation, failure, and uncertain early work.
4. Make room for analog serendipity and manageable learning difficulty.
5. Practice cognitive bootstrapping, using early search results to discover better terms, sources, and questions.
Make the balance deliberate. These practices treat intuition and algorithms as complements. A printed map can provide context while GPS pinpoints a location; a search engine can retrieve sources while a skilled searcher evaluates them. The goal is not slowness for its own sake. It is to preserve the human abilities that help people learn, notice surprises, and correct systems when their assumptions no longer fit.
The six practices form a practical design agenda: retain human judgment, preserve context, protect experimentation, and teach people to use information tools skillfully. Their deeper contribution is to make the costs of convenience visible before a system becomes dependent on it. The challenge is access. Time for exploration, skilled teachers, and physical alternatives are not equally available to everyone. A fair version of inspired inefficiency should be built into institutions, not left as a luxury for people who can afford to opt out.
Analysis
The book’s central contribution is to change the unit of analysis. Efficiency is usually presented as a ratio between inputs and outputs for a bounded task. Tenner asks what happens when the boundary expands across time, institutions, and human capability. A faster search may save minutes while narrowing discovery. Automated guidance may reduce wrong turns while weakening the skills needed when guidance fails. A medical record may standardize information while taking time away from the patient. The apparent paradox dissolves once costs shifted elsewhere, or postponed into the future, are counted.
This approach is strongest when it distinguishes transaction efficiency from innovation. Platforms can match supply and demand remarkably well without creating the kind of new capacity that continuous industrial processes once unlocked. It is also strongest in concrete cases where a proxy changes behavior, as with rankings, tests, and clinical alerts. These examples make a general systems insight tangible: optimization can produce local gains and global losses.
The argument should not be read as a case for slower tools or analog nostalgia. Tenner repeatedly acknowledges the real benefits of digital search, GPS, medical data, and automation. The more demanding question is how to preserve backup skills and adaptive capacity without wasting resources on every legacy practice. Not every inconvenience teaches, and not every human intervention improves a decision. Good design distinguishes productive difficulty from avoidable burden.
A useful extension is to evaluate a technology on four dimensions: immediate performance, long-term skill, resilience under failure, and distribution of benefits and costs. The first is easiest to measure, which helps explain why organizations overvalue it. The other three require broader evidence and judgment. The book’s cases suggest that systems become genuinely efficient when those dimensions are considered together, with people able to interpret data, challenge defaults, and recover when the optimized path breaks.
Review Summary
The Efficiency Paradox receives mixed reviews (3.28/5), with readers appreciating its central thesis that efficiency can create paradoxical inefficiencies, but criticizing execution. Many found the writing verbose, unfocused, and lacking cohesion between chapters. Positive reviews valued examples of "desirable difficulty," "productive boredom," and "waylosing," praising insights on how technology impacts learning, navigation, and medicine. Critics noted insufficient data, repetitive arguments, and misleading marketing about big data. Several reviewers felt the book reads like a literature review with interesting concepts but shallow analysis and excessive filler content diluting key points.
People Also Read
Glossary
Inspired inefficiency
Useful friction that builds capabilityThe deliberate preservation of effort, delay, or human participation when it supports a better long-term result. In Tenner’s account, an apparently slower practice can improve future efficiency by maintaining skills, allowing discovery, or making a system more adaptable. The term does not praise waste for its own sake; it asks which forms of inefficiency produce lasting value.
The Seven Deadly Sins of Efficiency
Seven ways optimization backfiresTenner’s taxonomy of how efficiency can undermine itself: counterserendipity, hyperfocus, self-amplifying cascades, skill erosion, perverse feedback, data deluge, and monoculture. Together, these describe how systems can suppress beneficial chance, narrow attention, amplify early choices, weaken human ability, distort incentives, overwhelm users with information, or become too uniform to respond to change.
Campbell's Law
Targets distort their measuresA principle about measurement and incentives: when a quantitative indicator is used to make consequential decisions, people have reason to manipulate the indicator, and the indicator can become a worse measure of the underlying goal. Tenner applies it to examples such as test scores, citations, customer reviews, and medical billing classifications.
Matthew effect
Advantage compounds through attentionA pattern in which people or works that have already gained recognition receive further attention and resources, while less visible competitors fall further behind. Tenner uses it to explain how rankings, citations, and recommendations can convert small early differences into large disparities. The resulting success may appear to reflect quality alone, even when cumulative exposure helped produce it.
Desirable difficulty
Effort that strengthens learningA learning principle in which a manageable challenge improves later recall or understanding by requiring deeper processing. Tenner discusses handwritten note-taking as an example: writing more slowly can force students to select and restate ideas rather than copy them word for word. The principle does not mean that confusing or poorly designed instruction is beneficial.
Perfect 5
Human-machine balance pointDavid Mindell’s term for a useful middle position on a spectrum between entirely human and entirely automated control. The exact balance varies by task. The central idea is that trained people should retain enough understanding and authority to guide or recover a system, while automation handles work it performs reliably.
Cognitive bootstrapping
Search that builds better searchesTenner’s method for learning an unfamiliar subject through successive searches. A first query identifies useful authors, concepts, or institutions; those discoveries then become search terms that reveal additional perspectives and sources. The method helps users build enough background knowledge to assess results, rather than treating the first page of search results as a complete map of a topic.
Waylosing
Productive, instructive disorientationTenner’s term for getting off the planned route in ways that enable discovery. Unlike aimless error, waylosing involves enough preparation and awareness to notice unfamiliar details, meet unexpected people, or develop a richer sense of place. It contrasts with navigation that treats reaching a destination by the fastest route as the only worthwhile outcome.
Hiding Hand
Obstacles reveal unforeseen capacityAlbert O. Hirschman’s concept for the way ambitious projects can begin because decision-makers underestimate the difficulties ahead. As obstacles emerge, people may discover skills, resources, and solutions they would not have mobilized if the full challenge had been clear from the start. Tenner uses the idea to explain why major innovations often require uncertain, inefficient experimentation.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.