Every test suite is a confession. It reveals exactly what a team was afraid of — and, more damningly, what never occurred to them to fear at all.
For decades, software testing was a human negotiation: which of ten thousand input combinations deserved an assertion, which edge case was worth an argument in a planning meeting. Then the machines arrived. Fuzzers that mutate inputs by the billion. Property-based testers that generate cases no engineer would imagine. LLM agents that write, repair, and file their own bug reports. The search space is being conquered faster than almost any other part of software engineering.
But search was never the hard part.
Drawing on the documented record — Therac-25, Knight Capital's catastrophic 2012 trading loss, Boeing's MCAS, the 2024 CrowdStrike Falcon outage, the XZ Utils backdoor — this book argues that every high-consequence failure in software history was a failure of judgment, not generation. A fuzzer finds a crash. It cannot tell you whether that crash matters.
Through the empirical literature on mutation testing, flakiness, and coverage metrics, this book traces exactly where automation stops and human responsibility begins — and why that boundary is more important now than ever.
Read it before your next release.
