I watched an engineering team spend three months building a perfectly tested, beautifully architected internal admin panel. During the same period, their authentication system had SQL injection vulnerabilities in production. Pristine code that nobody used. Dangerous code that everyone depended on.
This is the standard quality failure mode: treating quality as a uniform dial you turn up or down across the entire codebase. It's wrong. Quality is a finite resource you allocate -- lavishing it on the things that will destroy you if they fail, and deliberately withholding it from everything else.
The Allocation Decision
The question isn't "should we invest in quality?" It's "will this specific decision compound against us?"
Internal tools don't need tests. Your admin panel can have bugs. Your developer scripts can be hacky. Nobody will die. No customer will churn. No security researcher will write a blog post about your sloppy internal tooling.
Your authentication system, your data layer, your public API -- these are load-bearing walls. Cutting corners here doesn't save time. It borrows time at compounding interest. A SQL injection vulnerability in auth doesn't produce a bug ticket. It produces an incident report, a breach disclosure, and a conversation with legal.
Why Coverage Metrics Fail
Teams that optimize for code coverage spread quality uniformly across the codebase. Every module gets the same investment regardless of blast radius. The result: 95% coverage on an internal dashboard nobody uses, 60% coverage on the payment processing pipeline that handles every transaction.
The engineers who understand quality allocation ship faster and build more reliable systems simultaneously. They're not cutting corners -- they're choosing which corners to reinforce. An admin panel with rough edges and an auth system with zero known vulnerabilities is a better outcome than two systems with moderate quality.
Quality is a resource allocation problem. Aim it at the things where failure is catastrophic, and accept imperfection everywhere else.