Start free trial
Trending New Genres Authors
EnglishEnglish
EspañolSpanish
简体中文Chinese
繁體中文Chinese (Traditional)
FrançaisFrench
DeutschGerman
日本語Japanese
PortuguêsPortuguese
ItalianoItalian
한국어Korean
РусскийRussian
NederlandsDutch
العربيةArabic
PolskiPolish
हिन्दीHindi
Tiếng ViệtVietnamese
SvenskaSwedish
ΕλληνικάGreek
TürkçeTurkish
ไทยThai
ČeštinaCzech
RomânăRomanian
MagyarHungarian
УкраїнськаUkrainian
IndonesiaIndonesian
DanskDanish
SuomiFinnish
БългарскиBulgarian
עבריתHebrew
NorskNorwegian
HrvatskiCroatian
CatalàCatalan
SlovenčinaSlovak
LietuviųLithuanian
SlovenščinaSlovenian
СрпскиSerbian
EestiEstonian
LatviešuLatvian
فارسیPersian
മലയാളംMalayalam
தமிழ்Tamil
اردوUrdu
Searching...
SoBrief
Prediction Machines, Updated and Expanded

Prediction Machines, Updated and Expanded

The Simple Economics of Artificial Intelligence
AI makes prediction cheaper. What changes when businesses redesign decisions around it?
by Ajay Agrawal 2022 304 pages
3.86
3k+ ratings
Listen
Immersive
V2.3
Amazon Kindle Audible
Try Full Access for 3 Days
Unlock listening & more!
Continue

Key Takeaways

Treat AI as cheap prediction, not a machine that thinks.

The useful distinction. Prediction means using information already available to fill in information that is missing. It can estimate the future, identify what is happening now, or classify something hidden in existing data. A fraud model, for example, can infer that a current card purchase is suspicious without knowing the cardholder’s future. In this book’s economic lens, a prediction machine is a system that produces such estimates, not a general mind with human understanding.

Why it can feel magical. Google Translate became strikingly more fluent when it shifted toward machine learning, while fraud detection improved from catching roughly 98 percent of fraud to as much as 99.9 percent. That last improvement sounds small, but it cuts errors by about twentyfold. Prediction can perform tasks associated with intelligence without itself supplying human-like reasoning.

Analysis

This distinction is powerful because it replaces a vague debate about whether machines are intelligent with practical questions: what information is missing, what evidence can fill the gap, and what happens if the estimate is wrong? It also prevents capability demonstrations from being mistaken for general competence. A system that classifies images impressively may still lack common-sense understanding of what those images mean. The framework is most useful when treated as a tool for analyzing specific tasks, rather than as a final definition of every technology called AI.

When prediction gets cheap, redesign the business, not just the task.

Price changes behavior. The book applies a familiar economic rule: when something important becomes less costly, people use more of it, including in places where they had not used it before. Computers made arithmetic cheap, enabling digital photography. Machine learning makes prediction cheaper, supporting applications such as translating languages and navigating city streets, once too complex for lists of hand-coded rules.

Look for a strategic threshold. If Amazon could predict purchases accurately enough, it might send products before customers order them, changing a shop-then-ship model into ship-then-shop. That shift would also require new ways to collect unwanted items and manage returns. Cheaper prediction can raise the value of complements, such as data and delivery capacity, while reducing reliance on substitutes, such as human forecasting. The strategic question is what business becomes viable when uncertainty falls.

Analysis

The book’s strongest move is to treat AI as a change in the price of an economic input, not as magic that automatically transforms every company. That framing prompts leaders to look for thresholds: a prediction may need to become only slightly more reliable before a different operating model wins. The caution is that better forecasts do not guarantee better strategy. New logistics, customer consent, and operational capacity must also be in place, and those complements can be expensive or hard to coordinate.

Separate the forecast from the value judgment that makes a decision.

Prediction is not choice. Judgment is the work of deciding how valuable different actions and outcomes are. A weather model can estimate the chance of rain, but it cannot decide how much a person dislikes carrying an umbrella. Likewise, a credit card system can estimate fraud risk, but a company must weigh the cost of fraud against the frustration of blocking a legitimate purchase.

Make the objective explicit. A doctor deciding whether to treat a possible blood clot must consider not only how likely the diagnosis is, but also the consequences of acting wrongly. As prediction becomes cheaper, more situations become worth analyzing, which increases the need to define what counts as a good outcome. Those values may be hard to reduce to a simple formula, especially when people have different preferences or competing goals.

Analysis

Separating prediction from judgment reveals that technical accuracy cannot settle disputes about priorities. A model may estimate outcomes, while people still have to decide whose costs count, how to value a false alarm, and whether speed matters more than comfort. This connects AI design to the broader field of value-sensitive technology, where system goals encode human choices. The hard part is not merely finding an objective, but making disagreements visible rather than burying them inside a score or default setting.

Feed models the right data, not merely more data.

Three data jobs. The book distinguishes training data, which teaches a model patterns; input data, which lets it make a particular prediction; and feedback data, which helps assess and improve its performance. Cardiogram explored predicting irregular heart rhythms using wearable heart-rate data and information from about 6,000 users, including roughly 200 people already diagnosed. The example shows why labels matter: a stream of heartbeats alone cannot teach a model which patterns indicate a health problem.

Match data to the decision. Data collected for one prediction may be useless when making another. A hockey model trained on goals scored during a season cannot predict next season’s wins if those future goals are not yet known. More data can also have diminishing statistical returns, although a small improvement may still matter commercially if it makes a product more useful than competitors’ alternatives.

Analysis

Treating data as an investment sharpens the conversation from “How much data do we have?” to “What decision will this data improve, and can it be collected at an acceptable cost?” That distinction matters particularly in health, where always-on sensors can create privacy burdens alongside predictive value. The statistical gain from another observation may be small, while the business gain from crossing a reliability threshold may be large. Data value therefore depends on the decision it supports, not simply on volume.

Assign routine predictions to machines, and exceptions to human judgment.

Use different strengths. Machines can find complex patterns in large, repeated data sets, while humans can often reason from analogy when a case is rare or unfamiliar. In a breast-cancer image contest, a machine reached 92.5 percent accuracy and a pathologist reached 96.6 percent. Combining their assessments reached 99.5 percent because they made different kinds of errors. Prediction by exception means a machine handles routine cases and asks a person to focus on cases where its evidence is thin or uncertain.

Prediction can manage risk. Better forecasts may also reduce the need for buffers or insurance. Air Canada used cargo predictions to address unused capacity, with the system expected to reduce it by about a quarter. But forecasts need reliability information: a confident estimate based on missing or distorted evidence can make a decision worse, not better.

Analysis

This is a more useful model of human-machine collaboration than the simple contest of which side is smarter. It directs attention to error patterns, uncertainty, and the cost of human review. In practice, exception handling also needs clear escalation rules: a system that cannot recognize when it is outside its competence will not reliably summon help. Human oversight is not automatically protective either. Reviewers need time, authority, and relevant expertise, or the handoff becomes a ceremonial checkbox.

Redesign workflows around tasks before deciding which jobs disappear.

Start with the work. A workflow is made of tasks, and tasks contain decisions that may benefit from prediction. The AI canvas is a planning tool for making those parts explicit: what action is being considered, what must be predicted, what outcome matters, what information is needed, and how results can inform future predictions. It also forces an organization to explain what it means by “best,” a word that often hides competing goals.

Automation changes jobs unevenly. Ford expected computers to trim its accounts-payable staff by 20 percent. After redesigning the process, it reduced the department by 75 percent while making the work faster and more accurate. In other settings, tools remove one task and make others more important. Spreadsheets reduced bookkeeping calculations but increased the value of asking the right business questions.

Analysis

The task-level lens helps avoid two common mistakes: assuming that a tool will improve an unchanged process, and treating a job title as a single indivisible activity. It also exposes the organizational politics hidden in technical design. A system cannot optimize for “the best students” or “a successful hire” until the institution chooses what those outcomes mean. That clarification may be more disruptive than the software, since it can reveal that departments disagree about their actual objectives.

Let error costs, not accuracy alone, determine how much autonomy to grant.

Measure the stakes. Stakes are the losses that follow when an imperfect prediction guides an action. An incorrect Amazon product recommendation may disappoint a shopper but is unlikely to end the relationship. Facebook’s content moderation problem is different: a mistaken decision can expose many users to harmful material or wrongly suppress legitimate speech. In the book’s account, Facebook used AI to flag content but kept human moderators involved, including a large review workforce.

Shape the task around error. Spotify found that personalized playlists were not enough if users did not know what to play. Editors supplied context, such as music for singing in the car, while algorithms tailored selections to listeners. That design reduced the cost of a mismatch. For high-stakes decisions, leaders must consider false positives and false negatives separately, then decide whether prediction should advise, trigger, or fully control the action.

Analysis

Accuracy compresses several ethically different outcomes into one score. A false positive and a false negative can harm different people, at different scales, and with different routes to remedy. The relevant threshold is therefore partly a governance choice, not merely a technical setting. Human review can help, but it can also shift unpleasant work onto moderators or reviewers without giving them enough power. Good design makes the consequences and accountability for each error explicit.

Make learning a strategy, but choose carefully who bears its costs.

Learning needs exposure. An AI-first strategy prioritizes improving predictions, even when collecting useful experience may hurt short-term performance or customer satisfaction. Gmail’s suggested replies can tolerate frequent mistakes because users can ignore them. A driving system faces much higher costs when it is wrong, so learning from real-world use carries greater safety and reputational risks. Simulations can reduce those risks, but may not capture every condition a system encounters in practice.

Control the learning pipeline. Learning by using means that performance improves through repeated use and feedback. For a startup such as Ada Support, access to customer interactions mattered because those exchanges could improve its product. Apple, by contrast, emphasized privacy and kept some analysis on users’ devices, accepting limits on convenience and data sharing. An organization must choose when to deploy, what data it may use, and whether people retain skills while machines gain experience.

Analysis

Learning is not free experimentation. Someone supplies the data, tolerates early errors, and may absorb the safety or privacy costs while a product improves. This makes deployment timing a question of consent and risk allocation, not only competitive speed. The book’s examples also point to a human-factors concern: if automation removes routine practice, people may be less prepared for rare emergencies. A sound learning strategy develops the machine without quietly exhausting the human capabilities needed around it.

Audit AI for bias, brittle evidence, and risks that scale systemwide.

Neutral rules can discriminate. Research discussed in the book found that searches for Black-associated names were 25 percent more likely to display ads implying an arrest record than searches for white-associated names. Another study found that Facebook’s ad delivery could show fewer STEM job ads to women because men were cheaper to reach. Neither example required an explicit instruction to discriminate; outcomes emerged from the data and incentives.

Test what the model cannot explain. A prediction can reflect correlation rather than cause. eBay experiments found that search ads with a seemingly enormous measured return had little effect on sales for many existing customers. AI systems also face security threats: users can manipulate inputs, competitors can infer models through repeated queries, and malicious feedback can distort learning, as happened with Microsoft’s Tay chatbot.

Analysis

The central lesson is that auditing must examine who receives which outcomes, not just whether the model performs well on average. This aligns with the broader distinction between formal neutrality and substantive fairness: identical procedures can reproduce unequal access when their data or market context is unequal. Causal experiments can reveal whether an intervention actually changes behavior, while output audits can detect disparate effects. Neither is sufficient alone, particularly when users adapt strategically to the system.

Prepare for task displacement and concentrated returns, not a simple robot takeover.

Work changes in pieces. The book argues that AI is unlikely to end work altogether, but the transition can still be painful. Machines may take over prediction while people shift toward judgment, communication, or supervision. These roles may be more numerous yet pay less than the specialized jobs they replace. Because software scales quickly, displaced workers may face losses before replacement roles appear. Skill-biased technology can also concentrate gains among people and firms best able to use it.

Policy involves trade-offs. AI can raise productivity while increasing inequality, market concentration, or pressure on privacy. More data may improve personalization, but reduce control over personal information. Larger user bases can help models improve, but do not guarantee permanent monopoly. The book distinguishes these present economic effects from speculative fears about superintelligence, arguing that today’s prediction systems should be assessed on the decisions and risks they actually shape.

Analysis

The book resists both easy optimism and cinematic catastrophe by focusing on distribution, competition, and institutional choices. That is a useful corrective: aggregate productivity gains do not determine who receives them, and a long-run return to employment does not compensate someone for a difficult transition. The same framework suggests examining bargaining power, retraining access, and ownership of the systems that scale. Its reassurance about future employment is less decisive than its diagnosis of the near-term distribution problem.

Analysis

The book’s central contribution is a translation device. Instead of treating artificial intelligence as a mysterious new kind of actor, it asks readers to identify the economic input that has become cheaper: prediction. That move connects technical capabilities to familiar questions about costs, incentives, risk, and organizational boundaries. It also explains why the effects can be larger than the task initially automated. A better estimate can change which actions are feasible, which complementary resources become valuable, and which parts of a business belong inside the firm.

The framework is especially strong when it separates prediction from judgment. It makes clear that accuracy does not define success until someone specifies the objective and the cost of mistakes. That distinction supports practical analysis of product design, human oversight, and automation. The workflow lens extends the same logic to jobs: tasks change first, while occupations are bundles that can be recomposed around remaining human contributions.

The framework also has limits worth keeping in view. Treating AI primarily as prediction is analytically clarifying, but it can understate the importance of systems that generate content, communicate persuasively, or act with limited supervision. Even in those cases, however, the book’s questions remain useful: what information is being inferred, what objective governs the output, and who bears the consequences? A second limitation is that economic gains can appear cleaner in theory than in institutions. Power, labor bargaining, legal duties, and unequal access shape whether productivity benefits are broadly shared.

The book’s most durable lesson is therefore not that AI will replace people or enrich every firm. It is that cheaper prediction reorganizes decision-making, and those reorganizations distribute authority as well as value. Leaders should map the decision, make objectives explicit, test evidence, price errors, and redesign the surrounding workflow. Policymakers face the same task at a larger scale: preserve innovation while deciding how privacy, competition, safety, and income should be protected. The technology creates new options; institutions determine who can use them and who pays when they fail.

Summary in 30 Seconds
Machine learning lowers the cost of prediction, making new products and business models viable. A fraud system improved from catching about 98 percent of fraud to as much as 99.9 percent, cutting errors roughly twentyfold. Prediction estimates outcomes, but people must set goals and weigh error costs. Automate routine cases while routing uncertain or high-stakes ones to human judgment. AI reshapes tasks before whole jobs, while benefits and risks can accrue unevenly.
Contains spoilers

Last updated:

Report Issue

Review Summary

3.86 out of 5
Average of 3k+ ratings from Goodreads and Amazon.

"Prediction Machines: The Simple Economics of Artificial Intelligence" offers a clear, accessible introduction to AI's economic impact, focusing on its role in improving predictions. Readers appreciate its non-technical approach and real-world examples, though some find it repetitive or already outdated. The book's central premise—that AI's core function is prediction—resonates with many, while its insights on business strategy and societal implications are widely praised. However, opinions vary on its depth and relevance for those already familiar with AI concepts.

Your rating:
4.38
320 ratings
Want to read the full book?

Glossary

Prediction

Filling in missing information

The process of using information already available, often called data, to generate information that is missing. A prediction can concern the future, the present, or the past. In the book’s framework, machine learning is economically important because it makes this input to decision-making cheaper and usable in more settings.

Judgment

Valuing possible outcomes

The process of assigning value to different actions and outcomes so that a decision-maker can choose what to do with a prediction. Judgment identifies the relevant objective and weighs the rewards and penalties associated with correct and mistaken decisions. It is distinct from predicting what is likely to happen.

AI canvas

A tool for task analysis

A planning tool for examining whether an AI prediction can improve a particular task. It prompts a team to specify the action, prediction, judgment, outcome, and the information used to operate, train, and improve the system. By making those parts explicit, it helps reveal whether a prediction is useful and what organizational choices it requires.

Reward function engineering

Specifying values for outcomes

The work of determining how valuable or costly different outcomes are for an organization, given the predictions an AI produces. It can involve translating human priorities into rules that a system can follow, or deciding how people should act on predictions. Poorly specified rewards can lead a system to optimize a narrow measure at odds with wider goals.

Prediction by exception

Machines handle routine cases

A division of labor in which a prediction system handles frequent, familiar situations and directs unusual or uncertain cases to a human. The machine provides scale where historical data is plentiful, while human attention is reserved for cases where the model lacks evidence or context. The approach depends on detecting exceptions reliably.

Satisficing

Choosing what is good enough

A decision strategy associated with Herbert Simon, in which people settle for an acceptable option rather than search for an ideal one. Limited time, information, and mental capacity make exhaustive optimization difficult. The book uses this idea to explain why people build buffers and other compromises when prediction is weak.

AI-first

Prioritizing prediction quality

A strategy that places machine learning and prediction improvement ahead of some near-term goals, such as immediate customer satisfaction or operating performance. It requires resources for data collection and learning, and may involve deploying a system before it is fully mature. The choice trades short-term costs against the possibility of stronger future predictions.

Known knowns, known unknowns, unknown unknowns, and unknown knowns

A map of prediction limits

A taxonomy of situations that helps identify where machine predictions are reliable or vulnerable. Known knowns have sufficient data for dependable predictions. Known unknowns are recognized data gaps. Unknown unknowns are events outside prior experience. Unknown knowns are misleading patterns that look reliable but arise from hidden causes or changing circumstances.

FAQ

What's Prediction Machines about?

  • AI and Economics: Prediction Machines by Ajay Agrawal explores the economic implications of artificial intelligence, focusing on how AI transforms prediction into a cheaper and more accessible resource for businesses.
  • Core Component of AI: The authors argue that AI primarily enhances our ability to make predictions, which are essential inputs for decision-making processes across various industries.
  • Decision-Making Framework: The book provides a structured approach to understanding how businesses can leverage AI for better decision-making, emphasizing the importance of judgment alongside prediction.

Why should I read Prediction Machines?

  • Understanding AI's Impact: This book is crucial for anyone looking to grasp how AI is reshaping business strategies and decision-making processes in the modern economy.
  • Practical Insights: The authors, economists with extensive experience in technology, offer practical insights and frameworks that can be applied to real-world business scenarios.
  • Navigating Uncertainty: It helps readers navigate the uncertainties associated with AI adoption, providing a clear understanding of the trade-offs involved in leveraging prediction technologies.

What are the key takeaways of Prediction Machines?

  • Prediction is Central: The book emphasizes that the current wave of AI primarily enhances prediction capabilities, which are critical for effective decision-making.
  • Trade-offs in AI Adoption: Readers learn about the trade-offs associated with AI, such as the balance between speed and accuracy, autonomy and control, and data privacy.
  • Human Judgment Matters: While AI can improve predictions, human judgment remains essential for interpreting those predictions and making informed decisions.

What are the best quotes from Prediction Machines and what do they mean?

  • "AI enhances prediction": This quote highlights that AI's primary function is to enhance our predictive capabilities rather than replicate human intelligence.
  • "More data means less privacy": This statement underscores the ethical considerations and trade-offs that come with the increased reliance on data for making predictions.
  • "Weighing trade-offs is key": This emphasizes the importance of understanding the specific context and implications of decisions made in the AI landscape.

How does Prediction Machines define prediction?

  • Filling in Missing Information: The authors define prediction as "the process of filling in missing information," using available data to generate insights about unknowns.
  • Core to Decision-Making: Prediction is presented as a fundamental input into decision-making, allowing businesses to reduce uncertainty and make informed choices.
  • Economic Framework: The book uses an economic lens to analyze how cheaper predictions can lead to new business opportunities and strategies.

What is the relationship between prediction and judgment in Prediction Machines?

  • Complementary Roles: The book explains that while prediction machines enhance the ability to make predictions, human judgment is necessary to interpret those predictions and decide on actions.
  • Increased Value of Judgment: As prediction becomes cheaper and more accurate, the value of human judgment increases because it allows for better decision-making based on those predictions.
  • Decision Anatomy: The authors break down decision-making into components, highlighting that judgment, alongside prediction, is essential for achieving desired outcomes.

How does Prediction Machines address the trade-offs associated with AI?

  • Understanding Trade-offs: The authors emphasize that adopting AI involves trade-offs, such as balancing speed with accuracy and autonomy with control.
  • Framework for Evaluation: The book provides a framework for evaluating these trade-offs, helping readers make informed decisions about AI implementation in their organizations.
  • Real-World Examples: Through various case studies, the authors illustrate how different companies navigate these trade-offs in practice.

What are the implications of cheap prediction for businesses according to Prediction Machines?

  • Increased Usage of Prediction: As the cost of prediction decreases, businesses are likely to use predictions more frequently across various functions, from marketing to operations.
  • New Business Models: The book discusses how cheaper predictions can lead to the development of new business models, such as Amazon's potential shift from a shopping-then-shipping model to a shipping-then-shopping model.
  • Job Redesign: The authors suggest that businesses will need to rethink job roles and workflows to fully leverage the benefits of AI-driven predictions.

How does Prediction Machines suggest businesses can implement AI?

  • Identify Key Predictions: The book advises businesses to identify which predictions are most valuable for their operations and decision-making processes.
  • Invest in Data: Organizations should invest in collecting and managing data effectively, as high-quality data is essential for accurate predictions.
  • Integrate AI Tools: The authors recommend integrating AI tools into existing workflows and processes to enhance productivity and decision-making capabilities.

What challenges do organizations face when adopting AI according to Prediction Machines?

  • Data Privacy Concerns: The book highlights that increased reliance on data for predictions raises significant privacy concerns that organizations must address.
  • Resistance to Change: Organizations may face internal resistance when implementing AI solutions, particularly if these solutions disrupt existing workflows or power dynamics.
  • Skill Gaps: There may be a skills gap in understanding and utilizing AI technologies effectively, necessitating training and development for employees.

What is the AI canvas mentioned in Prediction Machines?

  • Framework for Decision-Making: The AI canvas is a tool introduced in the book to help organizations deconstruct tasks and identify where AI can be integrated.
  • Components of the Canvas: It includes elements such as prediction, input, judgment, training, action, outcome, and feedback, helping businesses assess AI's impact on workflows.
  • Practical Application: The authors illustrate the AI canvas with examples from various industries, demonstrating how it can be used to redesign workflows and improve decision-making.

What are the risks associated with AI as discussed in Prediction Machines?

  • Security Risks: The authors highlight the potential for data manipulation and hacking, which can compromise the integrity of AI predictions.
  • Quality Risks: The book discusses the risks of relying on AI predictions that may be based on incomplete or biased data, leading to poor decision-making.
  • Liability Risks: Organizations may face legal challenges if their AI systems produce discriminatory outcomes, emphasizing the need for monitoring and ethical compliance.

About the Author

Ajay Agrawal is a distinguished professor at the University of Toronto's Rotman School of Management, holding the Geoffrey Taber Chair in Entrepreneurship and Innovation and a professorship in Strategic Management. His academic work focuses on the economics of artificial intelligence, machine learning, and other emerging technologies. Agrawal is also known for his entrepreneurial initiatives, having co-founded NEXT Canada (formerly The Next 36) in 2010, an organization dedicated to fostering entrepreneurship and innovation among Canada's most promising young leaders. His expertise in AI and its economic implications has made him a respected voice in both academic and business circles, contributing to discussions on the future of work and technology-driven economic change.

Download PDF

To save this Prediction Machines, Updated and Expanded summary for later, download the free PDF. You can print it out, or read offline at your convenience.
Download PDF
File size: 0.41 MB     Pages: 21

Download EPUB

To read this Prediction Machines, Updated and Expanded summary on your e-reader device or app, download the free EPUB. The .epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.
Download EPUB
File size: 1.41 MB     Pages: 26
Want to read the full book?
Follow
Listen
Home
Swipe
Library
Get App
Try Full Access for 3 Days
Listen, bookmark, and more
Compare Features Free Pro
📖 Read Summaries
Read unlimited summaries. Free users get 3 per month
🎧 Listen to Summaries
Listen to unlimited summaries in 40 languages
—
❤️ Unlimited Bookmarks
Free users are limited to 4
—
📜 Unlimited History
Free users are limited to 4
—
📥 Unlimited Downloads
Free users are limited to 1
—
Risk-Free Timeline
Today: Get Instant Access
Listen to full summaries of 26,000+ books. That's 12,000+ hours of audio!
Day 2: Trial Reminder
We'll send you a notification that your trial is ending soon.
Day 3: Your subscription begins
You'll be charged on Nov 25,
cancel anytime before.
Consume 2.8× More Books
2.8× more books Listening Reading
Our users love us
1M+ readers
Trustpilot Rating
TrustPilot
4.6 Excellent
This site is a total game-changer. I've been flying through book summaries like never before. Highly, highly recommend.
— Dave G
Worth my money and time, and really well made. I've never seen this quality of summaries on other websites. Very helpful!
— Em
Highly recommended!! Fantastic service. Perfect for those that want a little more than a teaser but not all the intricate details of a full audio book.
— Greg M
Save 62%
Yearly
$119.88 $44.99/year/yr
$3.75/mo
Monthly
$9.99/mo
Start a 3-Day Free Trial
3 days free, then $44.99/year. Cancel anytime.
Unlock a world of fiction & nonfiction books
26,000+ books for the price of 2 books
Read any book in 10 minutes
Discover new books like Tinder
Request any book if it's not summarized
Read more books than anyone you know
#1 app for book lovers
Lifelike & immersive summaries
30-day money-back guarantee
Download summaries in EPUBs or PDFs
Cancel anytime in a few clicks

Why one summary?

Scanner
Find a barcode to scan

⭐
⭐
⭐
We have a special gift for you
Open
38% off
DISCOUNT FOR YOU
⭐
⭐
⭐
$79.99
$49.99/year
Only $4.16 per month
Continue
2 taps to start, super easy to cancel
Settings
General
Widget
Loading...