Notes
These notes provide sources for specific claims, statistics, and stories referenced throughout the book. Notes are organized by chapter and keyed to page phrases so you can find them without cluttering the text with superscript numbers.
Where examples are based on author experience or composites drawn from multiple interviews, that’s noted. All other claims point to their original sources.
Preface
“Seventy-two thousand messages a month”: Internal analytics, author’s social media engagement platform.
“Within 90 days, those workflows helped book $700,000 in new coaching contracts”: Attributed via the author’s CRM data. The figure represents booked contract value on coaching agreements signed during the 90-day window. Because most agreements were multi-month payment plans, cash collected within that period was lower.
“A consulting firm deployed AI to generate client proposals without review”: Composite example based on author observations across multiple client engagements.
Chapter 1: Why Your AI Strategy Is Backwards
“lawyers defending the Chicago Housing Authority filed a motion… Mack v. Anderson”: Jordan v. Chicago Housing Authority, Case No. 22 L 95, Circuit Court of Cook County, Illinois. Judgment entered January 15, 2025 after a jury found the CHA 100% responsible for the pediatric lead poisoning of two children, assessing $24.1 million between them. The March 17, 2025 post-trial motion described Mack v. Anderson as an Illinois Supreme Court decision; the CHA’s January 8, 2025 offer of proof had cited the same nonexistent case to the Illinois Appellate reports (“441 Ill. App.3d 819 (3rd Dist. 2021)”). At the July 17, 2025 hearing the defense characterized the problem as limited to “one case out of 50-some-odd case citations”; plaintiffs’ July 22, 2025 sanctions motion then documented 14 misrepresentations in the post-trial motion, 7 in the offer of proof, and 5 in a motion for extension—including a second fabricated case, First Chicago Bank v. Brandwein. Sanctions order entered December 5, 2025 (Judge Thomas Cushing): Larry Mason $10,000 personally, Goldberg Segalla $49,500. Danielle Malaty, who conducted the AI research, was not sanctioned in this order (she had not signed the filing and had been terminated in July). “they knew… ‘Fake Cases, Real Consequence’… adopted a formal AI policy five days before”: per the plaintiffs’ sanctions motion — Goldberg Segalla, “Fake Cases, Real Consequence: Misuse of ChatGPT Leads to Sanctions,” NYLitigator (NYSBA), October 2023 (the quoted warning is the article’s own language); internal policy prohibiting AI use without approval since July 2023; Artificial Intelligence Acceptable Use Policy implemented March 12, 2025 (motion filed March 17); D. Malaty, “Artificial Intelligence in the Legal Profession: Ethical Considerations,” September 4, 2024. Sources: Plaintiffs’ Motion for Sanctions (filed July 22, 2025); Chicago Sun-Times, December 9, 2025; ABA Journal, December 2025.
“a motion she drafted contained roughly a dozen citations ChatGPT invented—a different judge sanctioned her for those that July”: Calderon v. Dynamic Manufacturing, Inc., No. 2024 CH 09839, Circuit Court of Cook County, sanctions order July 16, 2025 (Judge William Sullivan). Malaty was sanctioned for a motion to dismiss and reply containing approximately 12 invalid ChatGPT-generated citations ($10 sanction plus $1,000 in opposing counsel’s fees). The order calls the “submission of twelve hallucinated citations… particularly egregious.”
“booked $700,000 in new coaching contracts”: See Preface notes above for the basis of this figure.
“roughly 500,000 calls and 50,000 automated text messages a month”: Author’s operating records from Pocket Your Dollars, where the author directed lead-generation operations prior to the company’s acquisition. Figures are monthly peaks at steady state.
“acquired by MediaAlpha, a public company, in a deal valued at up to $70 million”: MediaAlpha, Inc. (NYSE: MAX) Form 8-K filed February 24, 2022 (SEC EDGAR accession 0001818383-22-000021), disclosing an agreement to acquire substantially all assets of Customer Helper Team, LLC—the entity operating the Pocket Your Dollars business—described in the filing as “a leading provider of Medicare inquiries with deep expertise in social media marketing.” Management stated on that day’s earnings call that “the purchase price for the acquisition of CHT will be approximately $50 million in cash at closing plus up to an additional $20 million of contingent cash consideration based on CHT’s achievement of revenue.” The transaction closed April 1, 2022. The $70 million figure represents maximum total consideration; whether the full contingent amount was ultimately paid is not disclosed in public filings.
“95% of organizations investing in generative AI are getting zero return”: MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (Challapally, Pease, Raskar, Chari), July 2025. The report’s exact wording: “Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return.” Methodology (per the report): systematic review of 300+ public AI initiatives, interviews with 52 organizations, and surveys of 153 senior leaders, January–June 2025; the report labels itself preliminary and its figures “directionally accurate.” The finding drew methodological criticism after publication (e.g., that the related 5%-reach-production figure refers specifically to custom enterprise tools); the chapter quotes the report’s own top-line claim and characterizes the research honestly. Popularized by Sheryl Estrada, “MIT Report: 95% of Generative AI Pilots at Companies Are Failing,” Fortune, August 18, 2025.
“About 64% of CEOs adopted AI technology before understanding whether it would benefit their organization”: IBM Institute for Business Value, 2025 CEO Study, May 2025. The study found 64% of CEOs acknowledge that the risk of falling behind drives them to invest in some technologies before they have a clear understanding of the value they bring to the organization.
“In 1879, Edison demonstrated a practical light bulb… his Pearl Street station was selling electricity in lower Manhattan”: Edison’s practical incandescent lamp dates to 1879; the Pearl Street Station in lower Manhattan began commercial service September 4, 1882.
“The economist Paul David went looking for why… the unit drive… Productivity jumped”: Paul A. David, “The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox,” American Economic Review 80(2), 355–361, 1990. On the shift from central line-shaft (“group drive”) layouts to the “unit drive” (one electric motor per machine) and the resulting factory reorganization and delayed productivity gains of the 1920s, see also Warren D. Devine Jr., “From Shafts to Wires: Historical Perspective on Electrification,” Journal of Economic History 43(2), 347–372, 1983.
“the economist Robert Solow… ‘You can see the computer age everywhere but in the productivity statistics’”: Robert M. Solow, “We’d Better Watch Out,” New York Times Book Review, July 12, 1987, p. 36 (review of Cohen and Zysman, Manufacturing Matters).
“research from Aalto University found that higher AI literacy leads to more overconfidence”: Robin Welsch and colleagues, Aalto University; announced October 2025, published in Computers in Human Behavior. ~500 participants completed LSAT logical-reasoning tasks, half using ChatGPT; all users overestimated their performance, and those with higher AI literacy overestimated most—reversing the usual Dunning-Kruger pattern.
“once you’ve invested $180,000, killing the project feels worse than spending another $50,000”: Illustrative composite figures based on author’s observation of AI project cost escalation patterns.
David scenario ($127,000 across three AI initiatives; 89% false-positive rate; net 0.8% revenue loss): Composite example based on author’s observation of small-manufacturer AI adoption patterns. Figures are illustrative.
Rachel scenario (insurance chatbot trained on marketing materials; coverage misstatements to three policyholders) and Marcus scenario (AI-drafted competitive analysis with fabricated market-share figures): Composite examples based on author’s observations across client engagements. Figures are illustrative.
“Air Canada’s defense amounted to—in the tribunal’s words—suggesting the chatbot was ‘a separate legal entity’”: the quoted phrase is the tribunal member’s characterization of Air Canada’s submission in Moffatt v. Air Canada, 2024 BCCRT 149 (“In effect, Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions. This is a remarkable submission.”).
“Air Canada deployed a chatbot that told a grieving customer he could apply for bereavement rates”: Moffatt v. Air Canada, 2024 BCCRT 149 (British Columbia Civil Resolution Tribunal, February 2024). Confirmed via American Bar Association, “BC Tribunal Confirms Companies Remain Liable for Information Provided by AI Chatbot,” Business Law Today, February 2024.
“Amazon spent three years developing an AI recruiting tool before discovering it had learned to discriminate against women”: Jeffrey Dastin, “Amazon Scraps Secret AI Recruiting Tool That Showed Bias Against Women,” Reuters, October 10, 2018. Also covered: MIT Technology Review, “Amazon Ditched AI Recruitment Software Because It Was Biased Against Women,” October 2018.
Chapter 2: The Intern Model Explained
“72,000 customer messages per month” and “$700,000 in new contracts”: See Preface notes above.
“Content output tripled from 24 to 72 pieces per month”: Composite example based on author interviews with marketing teams implementing AI workflows, 2024-2025.
“Research time dropped from 40 hours to 16 hours per deal”: Composite example based on author interviews with financial analysts, 2024-2025.
Chapter 3: The CEO’s Information Problem
“large organizations run an average of 367 different software applications”: Forrester Consulting, Crisis of the Fractured Organization, commissioned by Airtable, 2022 (survey of 1,022 US/UK workers, fielded August–September 2022). The average large business uses 367 software applications and systems, with teams using an average of 45 daily and individuals averaging 52 weekly. Reported by CIO Dive and CDP Institute.
“An APQC survey of knowledge workers found that only 30 hours of a typical 40-hour week are actually productive”: APQC (American Productivity & Quality Center), Fixing Process & Knowledge Productivity Problems survey, announced November 9, 2021. Surveyed 982 full-time knowledge workers; found the average knowledge worker spends only 30 hours of a 40-hour week on productive work, losing roughly a quarter of the week to managing internal communication (3.6 hours/week), looking for needed information (2.8 hours/week), and unnecessary or unproductive meetings (2.2 hours/week).
“Worldwide email traffic has nearly doubled since 2015” and “over 120 emails per day”: Radicati Group, Email Statistics Reports (multiple editions). Radicati tracks worldwide email traffic growing from roughly 205 billion messages per day in 2015 to a projected 390+ billion by the mid-2020s—nearly double—and reports the average business user sends and receives 120+ emails per day.
“the average Teams user spends more than three times as much time in meetings as they did in early 2020”: Microsoft, Work Trend Index, March 2022 edition (“Great Expectations: Making Hybrid Work Work”). Reported a 252% increase in weekly time spent in meetings for the average Microsoft Teams user since February 2020—more than a tripling—alongside a 153% rise in the number of weekly meetings.
Chapter 4: Mapping Your Decision Landscape
“predicted 20-25 major decisions per week. The actual count: 132”: Composite example (Kenji scenario) based on author’s decision-mapping work with operations leaders.
“Response time to tenants improved 40%”: Composite example (Kenji scenario). Based on author experience with property management AI workflows.
“Triage time dropped from 8 hours per week to 2.5 hours”: Composite example (Kenji scenario).
“Call prep time dropped from 35 to 12 minutes” and “Quota attainment jumped from 108% to 127%”: Composite example (Aisha scenario). Based on author interviews with sales professionals implementing AI prep workflows.
“invested $45,000 in AI tool, unused for 6 months”: Composite example illustrating common pattern observed by author across multiple client engagements.
Chapter 5: Scoring Your Opportunities
“10 minutes/day x 250 workdays = 41 hours saved per year” vs. “2 hours/quarter x 4 quarters = 8 hours saved per year”: Author calculation for illustrative comparison. Math verified: 10 min x 250 = 2,500 min = 41.67 hours, correctly rounded to 41.
“Draft time dropped from about 45 minutes to about 10 per posting”: Composite example (Hugo scenario). Based on author interviews with HR directors.
“roughly 10 hours saved a week”: Composite example (Devon scenario). Based on author interviews with project managers.
Chapter 6: Choosing Where to Start
“Specific, concrete plans outperform vague intentions—research on implementation intentions has shown this repeatedly”: Gollwitzer, P. M., “Implementation Intentions: Strong Effects of Simple Plans,” American Psychologist, 54(7), 493-503, 1999. The implementation intentions framework demonstrates that specific if-then plans significantly increase goal completion rates compared to vague goal intentions; writing the commitment down is the author’s practical application of that specificity.
“4 hours saved across three meetings” by day five: Composite example (Farah scenario).
“140 tickets, 12 hours saved, zero quality issues” over 4 weeks: Composite example (James scenario). Based on author experience with IT support AI implementations.
Chapter 7: Your First Workflow
“Documentation time dropped from 45-60 minutes to 15-20 minutes per customer”: Composite example (Priya scenario). Based on author experience with customer success AI workflows.
“The team saved 10-15 hours monthly”: Composite example (Priya scenario).
Chapter 8: Role-Specific Applications
All workflow examples in this chapter are composite illustrations drawn from the author’s experience implementing AI workflows across multiple organizations and roles.
Chapter 9: Designing Inputs
“roughly a third cite output quality as their top AI barrier”: LangChain, State of Agent Engineering Report, 2025. Based on a survey of 1,340 professionals (November-December 2025). Confirmed: 32% cited quality as a top barrier, encompassing accuracy, relevance, consistency, and adherence to guidelines. Latency was second at 20%.
“roughly 80% of output quality comes from input quality”: Author’s rule of thumb from two years of workflow operation. Informal ratio, not a formal study.
Chapter 10: Output Quality
All examples in this chapter are composite illustrations based on author experience. No external sources cited.
Chapter 11: Risk Assessment
“Air Canada chatbot: Provided incorrect refund policy information”: Moffatt v. Air Canada, 2024 BCCRT 149 — the tribunal ordered Air Canada to pay C$812.02 total (C$650.88 in damages for negligent misrepresentation, C$36.14 pre-judgment interest, and C$125 in tribunal fees). See Chapter 1 notes for full citation.
“Amazon’s AI recruiting tool discriminated against women”: See Chapter 1 notes for full citation.
“In Mata v. Avianca (2023), attorneys submitted a brief citing cases that ChatGPT had fabricated”: Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023) (Castel, J., sanctions opinion and order, June 22, 2023). Attorneys Steven Schwartz and Peter LoDuca and the firm Levidow, Levidow & Oberman were jointly sanctioned $5,000 under Rule 11 for an affirmation citing six nonexistent cases generated by ChatGPT, and were ordered to notify their client and the judges falsely identified as authors of the fake opinions.
“manual data entry has a 3% error rate” and “AI-assisted verification can reduce errors to under 0.5%”: Peer-reviewed studies report manual single-entry error rates of roughly 1% for highly skilled operators to 4% for average operators, so 3% is a reasonable mid-range figure (Barchard & Pace, “Preventing Human Error: The Impact of Data Entry Methods on Data Accuracy,” Computers in Human Behavior 27(5), 2011). Verification workflows such as double data entry reduce error rates to roughly 0.3-0.5% or below (Goldberg and colleagues, JAMIA, 2008; a 2024 meta-analysis of clinical data processing found 0.29% for single entry vs. 0.14% for double entry). The sub-0.5% figure is documented for verification methods generally; the “AI-assisted” attribution in the chapter’s example benefit statement is illustrative rather than a specific study result.
Chapter 12: Data Access Decisions
“Samsung’s semiconductor division lifted its ban on ChatGPT… three separate times”: First reported by The Economist Korea (March 30, 2023, exclusive citing unnamed company insiders); Samsung has not publicly confirmed the incident details, hence “according to reporting.” The May 2023 temporary ban on generative AI tools on company devices was reported by Bloomberg from an internal memo and confirmed by Samsung in a statement to TechCrunch (Kate Park, “Samsung bans use of generative AI tools like ChatGPT after April internal data leak,” May 2, 2023).
“the EU AI Act”: Regulation (EU) 2024/1689 (Artificial Intelligence Act), entered into force August 1, 2024, with phased application beginning February 2, 2025.
All other examples in this chapter are illustrative composites and framework guidance.
Chapter 13: Career Protection
The “nightmare version” scenario (AI-generated product comparison shipped in a major campaign with a false competitive claim): Illustrative hypothetical, not a specific incident.
“Lawyers who use AI for legal research still bear professional responsibility”: American Bar Association, Formal Opinion 512, “Generative Artificial Intelligence Tools,” July 29, 2024. Confirmed. Addresses competence, confidentiality, communication, candor, supervisory responsibilities, and fees when using generative AI.
“CPAs who use AI for analysis still sign off on the accuracy of their work”: AICPA Professional Standards. The principle that professionals remain responsible for work product regardless of tools used is foundational to CPA licensing requirements. No single AI-specific guidance document supersedes existing professional responsibility standards.
“Healthcare providers who use AI for decision support still own the clinical judgment”: Established principle across medical licensing boards. The FDA’s framework for AI/ML-based Software as a Medical Device (SaMD) maintains that clinicians retain ultimate responsibility for clinical decisions, even when supported by AI tools.
Chapter 14: Calibrating Scrutiny
“Sustained vigilance degrades over time—psychologists have documented this since the 1940s”: Mackworth, N. H., “The Breakdown of Vigilance During Prolonged Visual Search,” Quarterly Journal of Experimental Psychology, 1(1), 6-21, 1948. Foundational study establishing the vigilance decrement. See also Warm, J. S., Parasuraman, R., & Matthews, G., “Vigilance Requires Hard Mental Work and Is Stressful,” Human Factors, 50(3), 433-441, 2008.
“Automation complacency—the tendency to reduce vigilance as systems prove reliable”: Confirmed by systematic review: Goddard, K., Roudsari, A., & Wyatt, J. C., “Automation Bias: A Systematic Review of Frequency, Effect Mediators, and Mitigators,” Journal of the American Medical Informatics Association, 2012 (PMC3240751). Source quote: “Automation complacency error rates for interruptive systems have been shown to increase if a DSS is highly (but not perfectly) reliable, leading to overtrust and complacency.” See also Parasuraman, R. & Riley, V., “Humans and Automation: Use, Misuse, Disuse, Abuse,” Human Factors, 1997.
Chapter 15: Building Review Into Workflow
“Mariana’s team of eight account managers used AI to generate weekly customer health reports… three months in, reports were going out unreviewed”: Composite example (Mariana scenario) based on author experience.
Chapter 16: Pattern Recognition
“According to a 2023 industry report, 73% of enterprises have adopted AI-assisted customer service”: This statistic is an intentionally fabricated example—it is the hallucinated citation the author almost shipped, used to teach readers what AI fabrication looks like. The report does not exist, by design. No correction needed.
Five AI error patterns (Hallucination, Outdated Information, Confident Uncertainty, Calculation Error, Context Misread): Author’s taxonomy based on observed patterns across AI workflow implementations.
Chapter 17: Workflow to Infrastructure
All organizational examples in this chapter are composite illustrations.
Chapter 18: Team Rollout
All team scenarios in this chapter are composite illustrations based on author experience with organizational rollouts. No external sources cited.
Chapter 19: Organizational Considerations
“Healthcare: HIPAA creates requirements around protected health information”: Health Insurance Portability and Accountability Act of 1996 (HIPAA), 45 CFR Parts 160 and 164. U.S. Department of Health and Human Services.
“Privacy regulations like GDPR affect how personal data can be processed”: Regulation (EU) 2016/679 (General Data Protection Regulation), European Parliament and Council, April 2016.
“Microsoft and LinkedIn’s Work Trend Index… 75%… 78%… 52%”: Microsoft and LinkedIn, 2024 Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part, May 8, 2024. Survey of 31,000 people across 31 countries, conducted February 15–March 28, 2024 by Edelman Data & Intelligence. The 78% BYOAI figure refers to the share of AI users (not all workers) who bring their own AI tools to work; 52% of AI users were reluctant to admit using it for their most important tasks.
Professional services firm AI adoption scenario: Composite example.
Chapter 20: Vibe Coding Introduction
The author’s pattern-detection tool and 12,000-message analysis are from personal experience. The ops lead’s CSV reformatting tool and other client scenarios are composite examples. Andrej Karpathy coined “vibe coding” in a February 2025 post on X.
Chapter 21: First Projects
All named scenarios in this chapter are composite illustrations; the author’s project stories are from personal experience. No external sources cited.
Chapter 22: Build vs. Buy
All company scenarios in this chapter are composite examples based on author experience. No external sources cited.
Chapter 23: Interactive Presentations
The author’s presentation stories (the SMS-question deck and the explorable follow-up) are from personal experience. Named scenarios are composite illustrations. No external sources cited.
Chapter 24: Master Prompt
All examples are illustrative composites and framework guidance. No external sources cited.
Chapter 25: Personal AI System
All examples are illustrative composites and framework guidance. No external sources cited.
Chapter 26: Skills Development
“fundamental interaction patterns persist across tools while specific feature knowledge expires”: Author analysis based on experience across multiple AI platform generations.
Chapter 27: Career Positioning
“when expert panels scored bariatric surgeons from operating-room video, the least-skilled quartile had nearly triple the complication rate and five times the mortality of the most-skilled”: John D. Birkmeyer and colleagues, “Surgical Skill and Complication Rates after Bariatric Surgery,” New England Journal of Medicine 369(15), 1434–1442, 2013. At least 10 surgeon-reviewers rated video of 20 practicing bariatric surgeons. Comparing the bottom quartile of peer-rated skill with the top: complication rate 14.5% vs. 5.2% (≈2.8×) and mortality 0.26% vs. 0.05% (≈5×), along with longer operations and higher reoperation and readmission rates.
“the measurable ‘CEO effect’ on company performance has been climbing for six decades”: Timothy J. Quigley and Donald C. Hambrick, “Has the ‘CEO Effect’ Increased in Recent Decades? A New Explanation for the Great Rise in America’s Attention to Corporate Leaders,” Strategic Management Journal 36(6), 821–830, 2015 (variance-partitioning across ~60 years and 18,000+ firm-years found the share of firm-performance variance attributable to the individual CEO rose substantially over the study period). On effect magnitude, see Donald C. Hambrick and Timothy J. Quigley, “Toward More Accurate Contextualization of the CEO Effect on Firm Performance,” Strategic Management Journal 35(4), 473–491, 2014 (the CEO effect approaches roughly one-third of performance variance in high-discretion settings, and is lower for large firms with entrenched structures). The chapter states the direction (rising over six decades), which is the robust finding; it does not put a single percentage on it.
“Most knowledge work roles aren’t being eliminated by AI—they’re being augmented”: Consistent with findings from the World Economic Forum, Future of Jobs Report 2025, which projects AI will create more jobs than it displaces. Also supported by McKinsey Global Institute research on workforce augmentation versus automation.
“Performance expectations are already rising”: Author observation consistent with Microsoft Work Trend Index findings and broad industry reporting on AI-driven productivity expectations in knowledge work roles.
“There’s a limited window where AI capability creates significant career differentiation”: Author analysis.
“This is the most valued skill among hiring managers”: Author observation. AI literacy has been cited as a top emerging skill in LinkedIn’s Workplace Learning Report (2024, 2025) and World Economic Forum workforce surveys, though no single survey uses this exact phrasing.
Chapter 28: Continuous Improvement
All observations in this chapter about model-generation turnover and practice decay are the author’s own, drawn from running the same workflows across three model generations.