AI Writes Your Tests. Here's What It Systematically Misses.

We ran Optinum against 16 real bugs from SWE-bench Verified and found that 62.5% of AI-written tests missed the exact failure class the bug belonged to — not random gaps, but the same structural blind spots.