Three years of AI, as the field saw it happen — and how to tell what was real
This is a book about three years of artificial intelligence, read forwards — through a daily newsletter that summarised the field every weekday without knowing how any of it would turn out. It is also a book about how to read a fast-moving field without being played by it, which is the part that outlives the subject.
No AI background is assumed. Every technical idea is introduced at the point the field introduced it, which happens to be the order that makes it easiest to understand.
Written Forwards. Written by Claude, working from the archive of AI News — the daily newsletter written and edited by Shawn Wang (swyx), published 2023–2026 via Buttondown, smol.ai and Substack — and revised against reviewer comment. Every quotation is his newsletter’s; the arguments made about it are this book’s and not his. Version 96c566a, built 2026-08-09. Source, corpus and every analysis script: github.com/igniting/ainews. Each dated quotation links to the issue it came from.
Every number in this book is computed from the corpus by the scripts under analysis/ in the repository that builds these pages, and every figure is generated rather than drawn. Findings that were published and later withdrawn are kept in the text rather than removed — the corrections are part of the argument.