Sources for Book 1: The Human Edge
About this list: it accompanies the English Amazon edition of Book 1 and gives readers current named links.
This page brings together bibliographic cards and verified direct links for Book 1. Cards follow the book's reading order, and their numbers match the source footnotes in the manuscript. Printed page numbers are intentionally omitted because they change with format and typesetting.
How to read the cards
- The description identifies the source and explains which fact, quotation, or qualification it supports.
- Open sources lists the named original publication, official document, independent confirmation, bibliographic record, or useful context.
- Gaps in numbering are expected: some numbers belong to glossary or explanatory notes that are not repeated here.
A linked source may open in the language in which it was originally published.
Introduction
Quotes and epigraphs
No. 1.
Tyler Cowen, remarks at the Sana AI summit, reported in Fortune, May 22, 2026. The epigraph uses the published English wording.
Open sources:
Research, reports, and data
No. 8.
Dell'Acqua et al., HBS Working Paper 24-013, and Ethan Mollick's public framing of the jagged frontier. Bounded study, used here as a general operating concept.
Open sources:
No. 9.
Dan Luu, “95%-ile isn't that good.” Personal essay used as an analogy, not population research about AI adoption.
Open sources:
No. 10.
Stanford Medicine overview of the Stanford-Harvard 2026 State of Clinical AI report. Used for the need for real-world evidence, accountability, and human judgment.
Open sources:
No. 14.
David Yang's “three crystals” framework, presented here as a practical entrepreneurial model rather than as a research finding.
Open sources:
Books and frameworks
No. 2.
In this book, AI means modern software that can read, write, analyze, and reason, especially models and agents made broadly available after 2022.
Open sources:
No. 12.
Michael Polanyi, The Tacit Dimension; David Autor, “Polanyi's Paradox and the Shape of Employment Growth.” The candidate avoids the absolute claim that an unstated rule is impossible for modern AI to learn.
Open sources:
- NBER — Polanyi's Paradox and the Shape of Employment Growth
- SSRN — Tacit Knowledge in Large Language Models
Industry, products, and cases
No. 3.
Sam Altman, “The Gentle Singularity,” 2025. Founder forecast about “intelligence too cheap to meter” and intelligence approaching the cost of electricity.
Open sources:
No. 4.
Microsoft corporate-agent announcements, 2024. Vendor framing, not a claim of universal reliability or adoption.
Open sources:
No. 5.
Andrej Karpathy, “Software Is Changing (Again),” 2025. Engineering framing for model-centered systems and human verification loops.
Open sources:
No. 6.
Peng et al., “The Impact of AI on Developer Productivity,” 2023. Controlled, narrow coding task; not evidence for every programming workflow.
Open sources:
No. 7.
OpenAI, GDPval, 2025. One-shot work-product evaluation across 44 occupations; inference estimates exclude human oversight, iteration, and integration.
Open sources:
No. 11.
Garry Kasparov, “The Chess Master and the Computer,” and Deep Thinking. The 2005 winners were two amateurs using three computers; coordination, not a single “better algorithm,” is the relevant lesson.
Open sources:
No. 13.
Frederick Reichheld, “Learning from Customer Defections,” Harvard Business Review, 1996. Supports retention economics without a universal acquisition-cost multiplier.
Open sources:
No. 15.
Newo.ai company materials and a sponsored implementation article. Used only as an operating illustration; no independent ROI or reliability claim is made.
Open sources:
Chapter 1
Research, reports, and data
No. 17.
OpenAI launched ChatGPT on November 30, 2022. The 100-million-user figure is a third-party estimate reported by UBS, not an OpenAI measurement.
Open sources:
- The Guardian — chatgpt million users open fastest growing
- OpenAI — chatgpt
- pcworld.com — midjourneys art goes live for everyone
No. 20.
a16z, “Welcome to LLMflation,” and Epoch AI, “LLM inference prices have fallen rapidly but unequally across tasks.” The rate varies by task, benchmark, provider, and threshold.
Open sources:
No. 22.
Anthropic Engineering, “How we built our multi-agent research system,” 2025. The roughly 4×/15× token ratios and 80% variance result are specific to its measured systems and BrowseComp analysis.
Open sources:
Books and frameworks
No. 16.
Andrew Ng's “AI is the new electricity” framing, 2016–2017. The metaphor describes cross-industry reach; it doesn't make AI economically identical to electricity.
Open sources:
- Fortune — Deep Learning Could Be the Next Internet
- Stanford GSB — Andrew Ng: Why AI Is the New Electricity
No. 25.
Paul David, “The Dynamo and the Computer,” 1990, with Karpathy's qualification of the electricity metaphor. Historical analogy, not a one-to-one forecast.
Open sources:
Industry, products, and cases
No. 18.
AutoGPT and BabyAGI illustrate the early agent loop; later product announcements illustrate computer, browser, and coding actions. Release claims don't prove reliable autonomy.
Open sources:
No. 19.
Microsoft agent announcements, 2024. Used only as vendor category framing.
Open sources:
No. 21.
Sam Altman, “The Gentle Singularity,” 2025. Founder forecast, retained with the historical “too cheap to meter” caveat.
Open sources:
No. 23.
Andrej Karpathy, “Software Is Changing (Again),” 2025. The computer-and-operating-system analogy is used as explanatory framing, not as empirical evidence.
Open sources:
No. 24.
Brett King, Bank 4.0: Banking Everywhere, Never at a Bank, 2018.
Open sources:
Chapter 2
Quotes and epigraphs
No. 32.
The maxim “Your margin is my opportunity” is widely attributed to Jeff Bezos. Its direct origin is not established, so it is presented as a business maxim rather than a verified quote.
Open sources:
Research, reports, and data
No. 29.
Simon Wu, “Distribution Is Oxygen. Endurance Is the Moat” (Cathay Innovation, December 2025). This is an investor's framework, not an independent market study.
Open sources:
No. 31.
Yi Zhou, “AI at the Edge of Transformation: Markets, Moats, and Momentum” (2026). This investor essay supports the customer-data-and-trust framework; it is not labor-market evidence.
Open sources:
Books and frameworks
No. 27.
Felix Oberholzer-Gee, Better, Simpler Strategy (Harvard Business Review Press, 2021). The value-stick framework connects willingness to pay and production cost.
Open sources:
No. 30.
Ben Thompson, “Aggregation Theory” (Stratechery, 2015). The framework explains why digital aggregators gain power through the customer relationship and demand.
Open sources:
Industry, products, and cases
No. 26.
Peter Thiel, Zero to One (2014) and “Competition Is for Losers,” The Wall Street Journal (September 12, 2014). The paired formula distinguishes unique value from competition on interchangeable output.
Open sources:
No. 28.
Warren Buffett, interview with the U.S. Financial Crisis Inquiry Commission (May 26, 2010). Buffett identifies pricing power as a central sign of a strong business.
Open sources:
Chapter 3
Research, reports, and data
No. 36.
Fabrizio Dell'Acqua et al., “Navigating the Jagged Technological Frontier,” Harvard Business School Working Paper 24-013, 2023.
Open sources:
No. 38.
World Economic Forum, Future of Jobs Report 2025. The 22 percent figure reports employer expectations about combined job creation and displacement through 2030.
Open sources:
No. 39.
U.S. Bureau of Labor Statistics, “Tellers,” Occupational Outlook Handbook, 2024–2034 projections. BLS projects a 13 percent decline and identifies online and mobile banking as important drivers.
Open sources:
Industry, products, and cases
No. 33.
John Maynard Keynes, “Economic Possibilities for Our Grandchildren,” first published in The Nation and Athenaeum in October 1930. Keynes described technological unemployment as a temporary phase of maladjustment.
Open sources:
- marxists.org — our grandchildren
- economicsnetwork.ac.uk — Economic Possibilities for our Grandchildren
No. 34.
James Bessen, Learning by Doing (Yale University Press, 2015), and “Toil and Technology,” IMF Finance & Development (March 2015). The ATM example explains task and role redesign; it is not a promise that teller employment will always rise.
Open sources:
No. 35.
OpenAI, “GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks,” 2025. The evaluation covers 44 occupations, nine industries, and 1,320 tasks. It measures one-shot task outputs rather than full jobs or unattended production systems.
Open sources:
No. 37.
International Labour Organization, “Generative AI and Jobs: A Refined Global Index of Occupational Exposure,” 2025. Exposure is more often associated with task transformation than complete job automation.
Open sources:
No. 40.
William Stanley Jevons, The Coal Question (1865). The rebound mechanism depends on demand expanding when each use becomes cheaper.
Open sources:
No. 41.
Ryan Roslansky and Aneesh Raman, LinkedIn commentary on an AI-enabled labor market. The three-basket adaptation in this chapter adds the author's economic and time-shift analysis.
Open sources:
Chapter 4
Research, reports, and data
No. 43.
Erik Brynjolfsson, Danielle Li, and Lindsey Raymond, “Generative AI at Work,” Quarterly Journal of Economics 140, no. 2 (2025). The study found an average 14 percent productivity gain for customer-support agents, with larger gains among less experienced workers.
Open sources:
No. 44.
Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” Stanford Digital Economy Lab, updated November 2025. The current version reports a 16 percent relative decline for workers ages 22 to 25 in the most AI-exposed occupations.
Open sources:
No. 47.
Cloudflare, company announcement, May 7, 2026. The workforce and AI-usage figures are company-reported. The builders, sellers, and measurers frame reflects management's explanation, not an independent causal estimate.
Open sources:
Books and frameworks
No. 42.
Aristotle, Nicomachean Ethics, Book II, on learning a craft through practice.
Open sources:
Industry, products, and cases
No. 45.
Matt Beane, The Skill Code (Harper Business, 2024). The chapter uses Beane's account of challenge, complexity or connection to the whole, and human relationships in skill formation.
Open sources:
No. 46.
Pia M. Orrenius and Madeline Zavodny, Federal Reserve Bank of Dallas, “AI and the Labor Market: Early Evidence,” January 2026. The authors describe aggregate effects as small and early, with hiring inflows a more plausible channel than broad layoffs.
Open sources:
No. 48.
Annelies Goger et al., Brookings Institution, “Generative AI, the American Worker, and the Future of Work,” April 2, 2026. Exposure estimates identify tasks open to change, not jobs certain to disappear.
Open sources:
Chapter 5
Quotes and epigraphs
No. 53.
Theodore Levitt, “Marketing Myopia,” Harvard Business Review (1960), popularizing the drill-and-hole illustration attributed to Leo McGinneva.
Open sources:
Research, reports, and data
No. 50.
Fabrizio Dell'Acqua et al., “Navigating the Jagged Technological Frontier,” Harvard Business School Working Paper 24-013, 2023. The experiment involved 758 consultants. Performance gains applied to tasks inside the tested AI frontier; outside it, AI users were less likely to reach a correct answer.
Open sources:
No. 51.
Harvard Business School Working Paper 26-036, field study of 244 consultants, 2025. The cyborg, centaur, and self-automator shares describe observed behavior in that study, not fixed personality types.
Open sources:
No. 52.
Perplexity and Harvard Business School, research-agent field study announced June 8, 2026. The comparison uses proprietary platform data and measured machine time and cost, not the full organizational cost of verification and deployment.
Open sources:
Industry, products, and cases
No. 49.
William Gibson's observation is widely cited from a 1999 interview and describes uneven adoption rather than a forecast about a specific technology.
Open sources:
Chapter 6
Industry, products, and cases
No. 54.
Herbert A. Simon, The Sciences of the Artificial, 3rd ed. (MIT Press, 1996). The passage links observed behavior to the structure of the surrounding environment.
Open sources:
No. 55.
OpenAI, “Harness Engineering: Leveraging Codex in an Agent-First World,” February 11, 2026. The article emphasizes environments, intent, feedback loops, and human steering.
Open sources:
No. 56.
Anthropic Engineering, “Effective Harnesses for Long-Running Agents,” November 26, 2025. The guidance focuses on context, incremental progress, environment design, and recovery across long-running work.
Open sources:
- Anthropic — effective harnesses for long running agents
- Anthropic — harness design long running apps 2
- GitHub — learn harness engineering
No. 57.
World Health Organization, “S.A.R.A.H., a Smart AI Resource Assistant for Health,” 2024. S.A.R.A.H. is described as a prototype digital health promoter. It doesn't establish that a conversational system can replace diagnosis or clinical care.
Open sources:
Chapter 7
Quotes and epigraphs
No. 68.
The line is widely attributed to Warren Buffett, but a reliable original source has not been established; it is used as an attributed saying, not data.
Open sources:
Research, reports, and data
No. 59.
Reporting by Euronews, CBC News, NBC News, and Rolling Stone, July 2025, on The Velvet Sundown, its audience, disputed identities, and public description as a synthetic project.
Open sources:
No. 60.
Lightcast, “Beyond the Buzz: Developing the AI Skills Employers Actually Need,” July 2025. The report analyzed 1.3 billion job postings and reported the stated salary premium.
Open sources:
No. 62.
World Economic Forum, Future of Jobs Report 2025. The figures are employer-survey expectations, not a prediction about one person's job.
Open sources:
- World Economic Forum — digest
- World Economic Forum — skills outlook 2
- reports.weforum.org — WEF Future Jobs Report
No. 64.
David Autor, “Polanyi's Paradox and the Shape of Employment Growth,” NBER Working Paper 20485 (2014).
Open sources:
No. 66.
OpenAI, GDPval (2025). Results vary by task and artifact type; the benchmark does not measure all professional judgment.
Open sources:
No. 70.
David Yan's three-crystal business framework, reported by Forbes in 2019. The claim that AI cheapened the middle crystal is the author's interpretation.
Open sources:
No. 72.
Onfido / Entrust, Identity Fraud Report 2024. Vendor data are presented as the provider's measurement, not a complete market estimate.
Open sources:
- entrust.com — onfido launches first fraud lab deepfakes
- thenextweb.com — deepfake fraud rise amid cheap generative
Industry, products, and cases
No. 58.
Immanuel Kant, Critique of Pure Reason, A133/B172. The wording follows a common English rendering of the passage on judgment and practice.
Open sources:
No. 61.
John E. DiNardo and Jörn-Steffen Pischke, “The Returns to Computer Use Revisited: Have Pencils Changed the Wage Structure Too?”, Quarterly Journal of Economics 112, no. 1 (1997).
Open sources:
No. 63.
Michael Polanyi, The Tacit Dimension (1966).
Open sources:
No. 65.
“Tacit Knowledge in Large Language Models,” SSRN preprint (2025). It is used as an argument about forms of knowledge, not as a settled law.
Open sources:
No. 67.
Steve Jobs, interview for PBS's Triumph of the Nerds (1996).
Open sources:
No. 69.
Herbert A. Simon, “Designing Organizations for an Information-Rich World” (1971).
Open sources:
No. 71.
Europol Innovation Lab, Facing Reality? Law Enforcement and the Challenge of Deepfakes (2022). Its often-cited 2026 figure was an expert forecast, not a measured share.
Open sources:
No. 73.
C2PA, Content Credentials Specification 2.4; European Commission guidance on AI Act Article 50 transparency obligations, applicable from August 2, 2026. Content provenance and proof of personhood address different questions.
Open sources:
Chapter 8
Quotes and epigraphs
No. 74.
Michel de Montaigne, Essays, Book I, Chapter 25, “Of Pedantry.” The English sentence is a compact rendering of Montaigne's passage.
Open sources:
Research, reports, and data
No. 78.
Hao-Ping Lee et al., “The Impact of Generative AI on Critical Thinking,” CHI 2025. The study analyzed self-reports from 319 knowledge workers across 936 tasks.
Open sources:
No. 80.
Michael Caosun and Sinan Aral, “The Augmentation Trap: AI Productivity and the Cost of Cognitive Offloading,” arXiv preprint (2026). The model examines how short-term productivity incentives can conflict with long-term skill formation.
Open sources:
No. 81.
Randazzo et al., HBS Working Paper 26-036, a field study built around a single experimental task involving 244 BCG consultants. The labels describe behavior in the observed session, not permanent personality types.
Open sources:
No. 84.
Drosos et al. (2025), research on embedded provocations intended to support critical and metacognitive thinking in AI-assisted work.
Open sources:
Books and frameworks
No. 79.
Wu et al. (2026), experimental work on AI assistance and reasoning practice. The claim is kept narrow: assistance design can affect learning; long-term effects remain an open question.
Open sources:
Industry, products, and cases
No. 75.
Warren VanderBurgh's “children of the magenta line” describes overreliance on flight automation. FAA and accident-safety guidance is used for the general human-factors mechanism, not to reduce any accident to one cause.
Open sources:
- Flight Safety Foundation — «Lost Skills»
- SAFE — «Automation Dependency: Children of the Magenta»
- Air Facts Journal — «Stepping down in automation»
No. 76.
Eleanor Maguire et al., “Navigation-related structural change in the hippocampi of taxi drivers,” Proceedings of the National Academy of Sciences 97, no. 8 (2000).
Open sources:
No. 77.
Betsy Sparrow, Jenny Liu, and Daniel M. Wegner, “Google Effects on Memory,” Science 333 (2011): 776-778.
Open sources:
No. 82.
Matt Beane, The Skill Code (2024).
Open sources:
No. 83.
Sycophancy is a model's tendency to reinforce a user's stated view instead of testing it firmly. The strength of the behavior varies by model, prompt, and evaluation.
Open sources:
Chapter 9
Quotes and epigraphs
No. 85.
Peter F. Drucker, “Managing Oneself,” Harvard Business Review (1999; reprinted 2005). Drucker recommends feedback analysis rather than relying on self-image.
Open sources:
No. 86.
Peter F. Drucker, “Managing Oneself,” Harvard Business Review (1999; reprinted 2005). Drucker recommends feedback analysis rather than relying on self-image.
Open sources:
Research, reports, and data
No. 88.
Randazzo et al., HBS Working Paper 26-036, a field study built around a single experimental task involving 244 BCG consultants. The observed session does not establish a long-term effect or a result for every kind of work.
Open sources:
No. 90.
OpenAI, GDPval (2025). Human-plus-AI speed estimates depend on task, model, review method, and study design.
Open sources:
No. 91.
World Economic Forum, Future of Jobs Report 2025. These are aggregated employer expectations, not a guarantee about net employment or one reader's profession.
Open sources:
Industry, products, and cases
No. 87.
Stanford Digital Economy Lab, “Canaries in the Coal Mine?” (2025/2026). The result is treated as an early labor-market signal consistent with pressure on some young workers, not proof that AI alone caused the change.
Open sources:
No. 89.
Michael Polanyi and David Autor on tacit knowledge and the historical boundary of automation. The chapter applies the idea cautiously to local context, verification, and responsibility.
Open sources:
Chapter 10
Quotes and epigraphs
No. 92.
Erich Fromm, To Have or to Be? (1976), Part I, Chapter VI.
Open sources:
Research, reports, and data
No. 93.
Rasmus Hougaard and Jacqueline Carter, More Human (2025). The book is used as a leadership argument about attention, wisdom, and care, not as labor-market data.
Open sources:
No. 94.
World Economic Forum, Future of Jobs Report 2025; Lightcast's 2025 job-posting analysis; and Stanford Digital Economy Lab's early work on young workers in AI-exposed occupations. The evidence is mixed and is presented as direction, not a universal forecast.
Open sources:
- World Economic Forum—Future of Jobs Report 2025
- Lightcast—Beyond the Buzz
- Stanford Digital Economy Lab—Canaries in the Coal Mine?
Chapter 11
Quotes and epigraphs
No. 95.
Amelia Earhart, “My Husband,” Redbook (1933), as cited in Mary S. Lovell, The Sound of Wings (1989).
Open sources:
Books and frameworks
No. 100.
Watkins, The First 90 Days. This chapter redefines the early win through the book's output/outcome distinction.
Open sources:
Industry, products, and cases
No. 96.
Brian P. Moran and Michael Lennington, The 12 Week Year (2013).
Open sources:
No. 97.
Michael D. Watkins, The First 90 Days (2003; updated 2013). Ninety days is used as a practical transition horizon, not a universal law of habit formation.
Open sources:
No. 98.
Ryan Roslansky and Aneesh Raman, Open to Work (2026), for the climbing-wall career image and a 30-60-90 framing. The plan in this chapter remains the author's own system.
Open sources:
- Microsoft WorkLab — The Essential Human Skills for AI Success: Focus on the 5Cs
- Microsoft Blog — Open to Work: How to Get Ahead in the Age of AI
No. 99.
Herminia Ibarra, Working Identity (2003).
Open sources:
Appendix 2
Industry, products, and cases
No. 101.
Anthropic, github.com/anthropics/skills and github.com/anthropics/knowledge-work-plugins, checked August 2026. These are primary vendor repositories, not independent evidence that every skill is effective or safe.
Open sources:
No. 102.
OpenAI Developers documentation for Codex skills and plugins, checked August 2026. A skill centers on a SKILL.md file; the current product documentation governs installation and distribution.
Open sources:
No. 103.
NVIDIA, github.com/NVIDIA/skills, checked August 2026. The repository describes signed entries, skill cards, and evaluation artifacts. Verification of publisher and integrity doesn't prove suitability for a particular use.
Open sources:
Appendix 3
Industry, products, and cases
No. 104.
Naval Ravikant, “How to Get Rich (without getting lucky),” and Eric Jorgenson, The Almanack of Naval Ravikant. The multiplication formula and transferability test in this appendix are the author's adaptation.
Open sources: