Code coverage percentages are one of the most misread metrics in software engineering. 85% coverage sounds healthy. But if 40% of your codebase is AI-generated tests covering AI-generated code that only validates the happy path, that 85% is telling you very little about actual defect detection.
A 2026 analysis found AI tools boost test coverage 40% while defect rates remain unchanged. The pattern is called coverage theater: tests that run and pass but don't meaningfully probe edge cases, error conditions, or the actual business logic. Standard coverage tools can't distinguish this from real coverage because they measure execution paths, not test quality.
GitIntel's coverage visualizer adds two dimensions to standard coverage reports. First, it overlays AI attribution on the coverage heatmap — so you can see at a glance whether uncovered code is untested human logic or untested AI-generated code. Second, it flags test files where both the test and the implementation are AI-generated without a human-authored test verifying the interaction, which is the pattern most associated with coverage theater.
Works with Istanbul/nyc (JavaScript), pytest-cov (Python), go test -cover, and Rust's built-in coverage tooling. Outputs HTML, LCOV, and Cobertura formats.