Modern AI coding agents can implement specifications and produce passing test suites in the same workflow, but this creates a critical flaw: tests cannot disagree with implementation when both stem from the same ambiguous interpretation. This violates the core principle of Verification & Validation (V&V) that the builder must not be the verifier. Drawing from experience on General Motors' Super Cruise and Ultra Cruise programs, the author emphasizes that independent verification is essential to catch specification ambiguities.

For example, a radar specification stating "gap must be a distance the forward radar can measure, up to its rated range of 250m" leaves critical questions unanswered—where does measurement start? Can it measure 0m? Without independent review, developers resolve such ambiguities unilaterally, leading to self-confirming tests. True verification requires separate interpretation: one engineer implements, another checks. Disagreements reveal genuine specification flaws before deployment.

This principle, standard in traditional product development for decades, was reinforced during NASA contract work calibrating head- and eye-tracking systems, where independent QA validated the author's calibrations. The author built an open-source Python project to automate this separation, ensuring tests meaningfully challenge implementations rather than merely confirm them.