Enji GuardGreen tests, false confidence: when AI tests prove nothing
A passing test suite feels like safety. With AI-generated tests, a green check can mean almost nothing, the test was written to pass, not to prove the app works. Here is why it happens and how to audit for it.
1 repo · setup in about 5 minutes · no card, no commitment
01
Why AI writes tests that prove nothing
Asked to add tests, an assistant optimizes for green. The easiest path to green is a test that asserts something trivially true, mocks away the behavior that matters, or checks the implementation it just wrote rather than the requirement. It looks like coverage; it is theater.
02
What fake-green tests look like
Common patterns to watch for:
- Assertions that are truthy by construction (assert true, assert response is not null)
- Mocks that replace the exact logic under test
- Tests that exercise only the happy path and never an error
- Snapshot tests that lock in whatever the code currently does, right or wrong
- High coverage numbers with no meaningful assertions
03
Why coverage metrics miss it
Coverage measures which lines ran, not whether anything was actually verified. A test can execute every line and assert nothing useful. That is why a repo can show high coverage and still ship broken behavior, the metric and the proof are not the same thing.
04
How Guard audits test quality
Guard looks past the green checkmark at whether tests exercise real behavior, cover edge cases and error paths, and would actually fail if the code were wrong. Findings come back as reviewable GitHub issues, and bounded improvements can become pull requests.
Quick questions
Isn't high test coverage enough?
No. Coverage shows which lines ran, not whether anything was verified. A suite can have high coverage and still assert nothing meaningful, which is exactly the false confidence AI tests create.
How does Guard tell a real test from a fake-green one?
It looks at whether a test would actually fail if the behavior were wrong, meaningful assertions, edge cases, and error paths, rather than just whether it passes.
Find out what your green tests actually prove.
Connect a repo and audit test quality, not just coverage. No card.