For developers shipping AI-written code
AI ships code that looks fine but would never survive review.
AI: Programming Like a God ~ The Heavenly Gates, reveals the layered stack of deterministic gates (tests, build, CI, AI reviews AI, etc.) that makes shipping rock-solid software the new standard.
Written in the open on Leanpub. Read who it is for before you buy.
Quality gates are the layers code must pass to reach production. A capable model games your gates: it rewrites the failing test until broken code goes green, then hands you slop that reads clean. The book builds a stack of gates one at a time against a real, public app, so you watch each one catch a live defect before the next goes in.
AGENTS.md and CLAUDE.md that load the right context and constraints into every AI session, so it starts inside your guardrails instead of outside them.
Compile-time guarantees and analyzers that fail the build, not the reviewer. The strongest green is the one the compiler enforces.
Tests, architecture fitness checks, and required checks that make merging below standard impossible, not merely discouraged.
A second model that reviews the first one's work against your standards, with verification loops that catch the confident-but-wrong output.
$ dotnet build && dotnet test
ok build 0 warnings (TreatWarningsAsErrors)
ok analyzers Roslyn + nullable clean
fail arch NetArchTest: Web references Infrastructure directly
fail mutation Stryker.NET score 58% below the 70% floor
merge blocked. the gate fails the build, not the reviewer.No human eyeballs are the last line of defense. The stack catches it first, every time. Every bug gets a gate.
No single gate is a proof. A capable AI can game any one of them. The book is honest about that: it teaches a layered stack of imperfect gates (defense in depth) sized to the blast radius, so the risk drops to a level your business can accept.
The running example is not invented. Gatekeeper is a real, public, MIT-licensed permissions app, built greenfield in the open in C#/.NET. It starts with no gates at all, and the book adds each one on camera. Every chapter ends as a git tag you can check out.
Tom Gilkison has written software for 25 years and works as a solutions consultant. In December 2025 he hit AI development and did not come up for air: 20 hours a day, seven days a week, for three months, building software, documentation, music, and anything else he could imagine. This book is the system he came up with along the way, the one that turns AI's output into rock-solid code instead of the default, out-of-the-box slop it consistently produces.
Read the opening chapters, then decide. The book is written in the open, and the price rises as chapters ship. Buy early to lock in today's price and get every update free.
Prefer to read in the browser? Read a free sample.