Two Types of AI Code Debt — And Only One Is Scannable

After scanning 200+ repos built with Claude Code, Cursor, and Copilot, we've identified two distinct types of debt: 1. Structural Debt — Scannable and Fixable Missing CI pipeline, zero test coverage, hardcoded secrets, unmanaged dependencies. This is measurable: 73% of AI-built repos have no CI pipeline 68% have zero tests 41% have hardcoded secrets These are binary checks. Either CI exists or it doesn't. Either secrets are in env vars or they're hardcoded. A scanner can catch all of this. 2. Comprehension Debt — Harder to Measure Code that works but nobody understands because nobody wrote it with intent. AI generates 200 lines where 40 would do. The abstractions are illogical. The reviewer's eyes glaze over. This debt compounds because each untested module interacts with other untested modules. The failure modes multiply faster than the code volume. The Gap Most teams don't realize they have both types until the first production incident. By then, the structural debt has made the comp