Product Quality in the AI SDLC: Why the Auditor Cannot Be the Creator

In the AI Software Development Life Cycle (SDLC), writing code is no longer the primary constraint. Instead, verifying the quality of the code has taken its place. Artificial agents can generate code much faster than any review or testing process can handle, and they are unable to determine if the outcome is satisfactory. Consequently, the auditor cannot be the creator, as the person who writes the test suite merely tests what they originally envisioned. This issue affects both human creators and AI agents.

The bottleneck has shifted from writing code to verifying it. Every process previously designed to manage the complexity of writing code now focuses on ensuring the code's quality. The assumption that writing code is the most challenging aspect of the process has become obsolete, as demonstrated by metrics such as pull requests merging without any review, both human and AI.

With pull requests merging without review, Faros AI's AI Engineering Report 2026 found that 31.3% of PRs merged without review, and the median time for review increased by 441.5%. Since reviewers cannot keep up with the volume of code, more code reaches production without being examined. The cycle of shipping code, finding issues manually, writing tests for the discovered problems, and experiencing those tests breaking upon subsequent refactoring continues unabated.

In this scenario, hiring additional Quality Assurance (QA) personnel does not resolve the loop. It merely adds more staff to the problem. Therefore, the bottleneck has moved to verification and judgment. The question has shifted from "Can we build it?" to "Does it work, does it hold together, and is it good enough to ship?" These are product-related questions that lack predetermined answers. The mechanics of the cycle are explained in more detail in the article "Past the Bottleneck."