The launch story
How this book was built
I finished and published From Archivist to Architect by running a coordinated team of Claude Code sessions. This page is the honest record: what the machine did, what I did, and what went wrong on the way.
The manuscript was not new. I am Jovan Smith, and I drafted it into May 2026, when it was reviewed as publication-ready in substance. Then I stalled. For two months the draft sat in Google Drive while the launch work waited for a stretch of open time that never arrived.
The restart began with an inventory, not with writing. On Saturday, July 18, 2026, the sessions recovered the draft from Google Drive into a fresh private repository, along with a full accounting of what existed, what was missing, and what was already good enough. The finding: the remaining work was launch execution, not drafting.
The build
On Tuesday, July 21, 2026, I uploaded the finished book to Amazon KDP. In between came the coordinated part. I ran the Claude Code agent sessions the way a site lead runs a crew: parallel sessions on planning, manuscript passes, figures, metadata, and quality checks, with every ruling captured in a decisions ledger the moment it landed. Nothing had to be re-litigated at 3am. The final edition is 10 chapters and 21,546 words, with 18 figures built in Canva.
The overnight quality pass treated the book like production software. A proofread logged 19 fixes. A cover-to-cover reader pass flagged 28 places where a real reader would stall, each one reviewed. Every figure got its own visual QA check. The finished EPUB validates clean in EPUBCheck (the EPUB format validator).
Every merge along the way, and the publish click itself, was mine.
None of this was the machine writing a book by itself. It was one operator who knows the material using a powerful tool at full speed. The judgment stayed human at every gate. That is also what the book argues. Chapter 10 documents the exact workflow that produced it, including a section titled "The Proof You Are Holding."
What went wrong
Plenty. A 10pm publish target slipped, and the finish became an overnight session. A duplicate pull request had to be closed as superseded. A late change to the merge order could have scrambled the final build. The decisions ledger and the human review gates caught it before anything shipped wrong. The point of the system was never zero mistakes. The point was making them visible while they were still cheap.
Read more
The full build story, decision by decision, is written up in the companion repo: read the complete how-this-was-built write-up.