Notes & Sources

A Note on the Stories in This Book

The firsthand stories in this book are drawn from the author’s own work as a developer, architect, adviser, and founder, and from founders and teams he has worked with directly. Companies, clients, and individuals have been generalized or anonymized where confidentiality or courtesy required it, and identifying details have been removed rather than replaced. No customers, conversations, results, or figures were invented. Where numbers appear, they are stated at the level of precision the author’s records support and are often rounded; where a figure was estimated or modeled rather than measured, the text says so. Percentages described as relative improvements are relative, not percentage-point, changes.

Scenarios introduced with words such as suppose, imagine, or hypothetical are constructed to make a piece of reasoning concrete. They are not disguised case studies and should not be read as events that occurred.

The book’s tools, including the Commercial Failure Diagnostic, the Paid Problem Test, the Build Permission Checklist, the Commitment Evidence Ladder, the Buyer Stress Test, the First Customer Path, and the Payer Delta, are the author’s own. They draw on broad principles common to the practice of selling software, such as the importance of outcomes over feature lists, the priority of behavioral over stated evidence, and the discipline of diagnosing before scaling, without reproducing any other author’s named framework.

Introduction

The description of coding as a major AI use and of agentic tools performing longer sequences of work draws partly on Anthropic’s Economic Index research. Anthropic’s “AI’s Impact on Software Development” (April 28, 2025), based on 500,000 coding-related Claude interactions, reported that 79 percent of Claude Code conversations in its sample were classified as automation rather than augmentation. Its “New Building Blocks for Understanding AI Use” report (January 15, 2026) reported that computer and mathematical tasks remained about a third of Claude.ai conversations and nearly half of first-party API traffic in its November 2025 sample. These are vendor-generated observational data about Claude usage, not market-wide measures of developer productivity.

The productivity comparison rests on controlled studies with different designs and populations. A controlled experiment by Sida Peng, Eirini Kalliamvakou, Peter Cihon, and Mert Demirer, “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot” (arXiv:2302.06590, February 2023), found that developers given an AI pair-programming tool completed a defined implementation task substantially faster than a control group. A randomized controlled trial by Joel Becker, Nate Rush, Elizabeth Barnes, and David Rein, “Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity” (arXiv:2507.09089, July 2025), found that experienced open-source developers working on tasks in their own mature repositories took longer when allowed to use early-2025 AI tools, while believing afterward that the tools had made them faster. On February 24, 2026, METR reported that a follow-on experiment with newer tools had become difficult to interpret because selection effects made the current productivity estimate unreliable. The point made in the Introduction is deliberately narrower than any one result: the effect depends on the tools, tasks, developers, and study design, and none of these studies establishes whether buyers will pay for what gets built.

Chapter 3

The account of a free-heavy SaaS application for technical professionals, the introduction of a paid plan, and the subsequent narrowing of the paid offer to users doing paid client work is from the author’s direct involvement. The recurring-revenue figure is rounded and reflects the period described.

Chapter 4

The integration story, including the proposed annual agreement, the conditional-commitment question, and the later customers who committed before the integration was built, is from a team the author worked with. The contract and first-year revenue figures are rounded.

Chapter 5

The landing-page account, including the five buyer interviews and the use of small paid-traffic tests as an experiment rather than a rescue, is from the author’s direct work. The conversion improvement and later recurring-revenue milestone occurred over the periods stated; the author does not attribute the whole of the later growth to the changes described, and other changes to the business occurred during that time.

Chapters 6 and 7

The account of targeting CTOs before discovering that an operations director owned the consequence, and the account of an internal operations application that gained adoption only after the rollout was narrowed around one high-volume workflow, are from the author’s direct work. Pilot and contract figures are rounded, and the estimate of staff time consumed by the reporting process in Chapter 6 is approximate.

Chapter 15

The enterprise AI deployment described, including the narrowing of the first release and the documentation of controls, human review, and ownership, is drawn from the author’s enterprise work and anonymized. The figure given describes a business case rather than independently measured savings, and the text is worded accordingly.

Chapters 17 and 18

The founder with a large public audience and no revenue, and the consultant path that produced the first paying organizations, is from the author’s advisory work. Impression figures are approximate, and the customer count, recurring-revenue figure, and share of customers originating from referral relationships are rounded.

Chapters 20 and 24

The founder building an AI application for agencies, the two-week pause in outreach, the twelve interviews, the setup-plus-subscription offer, and the business a year later are from the author’s coaching work. Deal values and the later customer count and recurring-revenue figure are rounded. The pricing story in Chapter 24, in which a $29-per-month utility was repackaged as a structured analysis offer, is from a company the author worked with; the average-revenue comparison covers the year following the change, during which other aspects of the business also changed.

Chapter 23

The account of a company with strong acquisition and weak retention, the reconstruction of the first fourteen days after purchase, and the onboarding redesign are from the author’s advisory work. The retention improvement is a relative comparison against the prior cohort. The company’s later annual-recurring-revenue milestone occurred within the period stated and reflects many changes beyond the onboarding work; the text does not attribute it to that work alone. The retention-math examples in the chapter are illustrative back-of-the-envelope calculations using constant-rate assumptions; the text distinguishes logo churn from MRR churn and recommends cohort analysis when customer sizes, expansion, contraction, or churn behavior vary.

Practical Tools

The appendices restate tools introduced in the chapters indicated. They are designed for a first pass in under twenty minutes and are intended to produce a decision or a next test rather than a document. None requires any software to use.