Currently writing
Sovereign AI: First Principles (working title)
Sovereign AI as a discipline, not a marketing term. Written from a
desk running a 119B-parameter model. Free to read, pay-what-it-was-worth.
A hybrid book: a principles framework, a narrative engineering log,
and an operator-pattern appendix. The narrative spine is the
timeline:
what was running, what broke, what got
merged upstream,
what got deferred and why. The principles layer extracts the rules
the operating log keeps confirming; the appendix is the pattern
library.
The closest current preview of how the book is organised is the
Sovereign AI Stack reference architecture:
sixteen sections layered from hardware up to operator-discipline,
the same arc the book follows.
Voice and discipline are spelled out in
The Engineering Honesty Manifesto:
six rules every chapter inherits. No fabricated numbers, no
anonymous-expert citations, no stealth-edited claims.
Distribution model: no-ISBN-first via Leanpub,
Gumroad, direct Lightning payment, and print-on-demand fallback.
CC BY-SA licensed, no-KYC, value-for-value priced. The shape of the
publishing economics matches the shape of the stack the book
describes.
Status: drafting against a six-month target. No pre-order page yet.
Sample chapter will land at /books/sample/ when it is
ready to read; subscribe via
RSS
if you want to know. The
no-newsletter decision
applies here too.
Recommended reading
Each book here is one the operator has finished and would re-read. No
"books I plan to read": that list is private and notoriously dishonest.
-
The Pragmatic Programmer
Andrew Hunt & David Thomas
Still the single best book on the operator mindset. The DRY principle and "broken windows" land harder once you've maintained a system long enough to live with your own past shortcuts.
-
Designing Data-Intensive Applications
Martin Kleppmann
Distributed-systems fundamentals presented without hype. The bits on replication and consistency translate directly to thinking about model-weight provenance and cache invalidation across an inference stack.
-
Site Reliability Engineering
Google SRE team
The free O'Reilly version is enough. Read it for the SLO discipline and the on-call psychology: both transfer to running a one-person AI stack where you are also the on-call.
-
The Soul of a New Machine
Tracy Kidder
Hardware bring-up narrative from the late 70s. The cultural pattern of small teams shipping ambitious silicon is the closest analog to the self-hosted-AI moment we're in.
-
The Sovereign Individual
James Dale Davidson & William Rees-Mogg
1997 prediction-essay that ages strangely well on the political economy of digital sovereignty. Read with calibration: the policy predictions are hit-and-miss, the technological direction-of-travel is uncomfortably accurate.
-
The Bitcoin Standard
Saifedean Ammous
Monetary history through a hard-money lens. The "why a no-KYC stack at all" question gets a structural answer here rather than a vibe-based one.
-
Layered Money
Nik Bhatia
The mental model of money as layers (gold → bank deposits → Eurodollars → CBDC → Bitcoin → Lightning) is the cleanest framework for explaining why the Lightning Address in the footer matters.