Sources and further reading
The standards, tools and sources the article draws on, by chapter, with what each one supports. The project's own figures come from its repository and are given as of October 2026. The product, its clinical domain and its market are deliberately left unnamed.
Prologue · The gate before the bet
There's a new kind of coding I call "vibe coding"
The post that named vibe coding: accept what the model writes and "forget that the code even exists". The practice the prologue argues against.
The healthcare data exchange standard behind the three server implementations of the first experiment.
01 · The claim, and what it is not
IEC 62304:2006+AMD1:2015, Medical device software: software life cycle processes
The standard for medical device software, and its safety classes. Class C applies to software whose failure could cause serious injury or death, the class the device is developed to.
The subscription behind the €20 a month figure. Prices as of October 2026.
02 · The barriers AI lowers
Noriaki Kano, Nobuhiku Seraku, Fumio Takahashi and Shinichi Tsuji, "Attractive quality and must-be quality"
The origin of the Kano analysis used to sort the product's features into those customers take for granted, those that satisfy more the more of them there are, and those that delight because nobody expected them.
04 · What BMAD brings to Claude
BMAD Method and its documentation
The roles, workflows, story preparation, three-layer code review, change proposals and party mode described in chapters 04 and 05.
The agent that reads the code, writes files, runs the tests and commits its changes, discussed in "An agent from day one?".
The other methods and tools I read about before choosing BMAD. I did not test them.
06 · The QMS as backbone
ISO 14971:2019, Medical devices: application of risk management to medical devices
The risk management standard behind the risk register and the risk management file.
IEC 62304:2006+AMD1:2015, as in chapter 01
The life cycle the software development file, traceability and verification records follow.
The plugin that sets up the quality system and wires it into BMAD's workflows, with its end-to-end tests (Appendix F).
Appendix A · How long would the configuration tool have taken by hand?
Software Cost Estimation with COCOMO II
The model behind the first estimate: its effort and schedule equations, and the settings for reliability, documentation, team and scale factors. The COCOMO II.2000 calibration is the one used.
The model's home today, kept by the research centre named after its author: an overview of the model, its submodels and what it is meant for.
COCOMO II, Constructive Cost Model
An online calculator by one of the book's co-authors, for readers who want to rerun the estimate with the size and settings listed in Appendix A, or with their own.
Ideal days, and why they differ from elapsed working days: the basis for converting the story estimates of the second method.
Les salaires informatique et électronique
Gross annual salaries in France by role, seniority and zone: Paris area, large cities, other regions. The basis of the salaried costs in Appendix A, and one survey among several: France has no official reference for salaries.
Elsewhere on this site
How I followed AI up to the launch of ChatGPT, where the prologue picks up.
GenAI Beyond the Hype, chapters 02, 03, 04 and 05
The cost of AI moving from the price to the volume, and the volume and pricing test (chapter 02); the crawl, walk, run, fly stages and the conditions for Fly (chapter 04); no claimed result without a baseline (chapter 01).
Beyond the Demo, chapters 05 A business case built, not proven and 06 Bringing control back: a specific Definition of Done
The business case built step by step, and when stopping is the business case doing its job (chapters 01, 02 and 10); the Definition of Done every story must clear (chapter 06), and that the product, and now my own tools, are held to (chapter 07).
Building It: turning governance into a product, including The walking skeleton and the roadmap
The walking skeleton as the first stage of a roadmap (chapter 03).
Domain-Driven Design and Scrum: Drawing the Lines, including DDD in brief and Bounded contexts as risk boundaries and A language you can prove
Core, supporting and generic subdomains (chapter 02); the boundary between two bounded contexts as the place where the risk sits, tested first by the walking skeleton (chapter 03); bounded contexts, the core domain and the ubiquitous language that keeps requirements, tests and code aligned, applied to the product's boundaries (chapter 08).
Agile Planning, including Start with a walking skeleton, Grow the skeleton and Ranges, buffers and fixed dates
The Kano model for ordering a backlog (chapter 02); the walking skeleton first, and how it grows (chapter 03); planning horizons and forecasts stated as ranges (chapter 10).
Scrum in Name Only, including The visible tip, Stories without increments, Who fits, What Scrum doesn't cover, Weaving architecture back in and Rebuilding from the bottom up
Architecture that is risk-driven and continuous, just enough and just ahead of need (chapter 03); Scrum as purposefully incomplete, with regulatory deliverables and the longer horizon to add deliberately (chapter 04); inspection only as good as the openness and courage behind it, and making it safe to report bad news early (chapter 05); the cross-functional team and who fits in it (chapter 11).