Or So I Was Told

Your tests pass. The pull request merged. The issue's closed. The agent says it's done.

None of that means your software works.

Every signal you've been trained to trust is a proxy. A green light standing in for something you can't see in the moment. And a proxy can be perfectly green while the thing underneath is broken. When a machine writes the code, that gap becomes the most dangerous place in your workflow, because the machine will make its proxies green whether the work is right or not. It has no stake in the outcome and no memory of the last time it was wrong.

*Or So I Was Told* is a field manual for the job that's left when the machine does the typing: verifying what it actually built. Not "did it run." Not "did the agent say it checked." How do you *know*?

This isn't a survey of AI coding tools, and it isn't a pile of tips. It's built from a real, instrumented development pipeline: 199 shipped issues, a cost ledger, and a log of every failure that taught the author something, where the claims are grounded in incidents that actually happened, with the receipts attached. And it practices what it preaches: the book was AI-drafted under the author's direction and put through the same verification gauntlet it teaches — a voice pass, a fact-and-plausibility pass, an editor, and a human who caught what the machines missed. The making of the book is the first case study in it. (The full story is in the preface. Read that first.)

You'll learn to tell the difference between a check that means something and one that only looks green: deterministic gates the machine can't argue with, verifying against real behavior instead of mocks, why a security pass and a correctness pass are different jobs, how a weak check quietly trains bad behavior, and how far you can safely let an agent run on its own.

This is a book in progress. You're reading it as it's written which, for a book about not trusting things that only look finished, is the only honest way to sell it. Buy now, pay less, and get every new chapter free as it clears the gauntlet. Early readers are part of the verification loop: catch something wrong, and you might end up in the log.

This is Book 4 in the *Outgrowing the Vibe Code* series: the verification layer, after harness engineering, spec-driven development, and keeping AI on your terms.

For developers who use AI to write real software and are done taking its word for it.