Find the technical debt AI-generated code introduces

The AI debt signals Guard watches

Executive Summary

• Guard monitors the debt fast AI-assisted development leaves behind: duplicated patterns, weak verification, dependency drift, lost product intent, and architecture that becomes harder to change.
• It is the gap between code that works today and code your team can safely change tomorrow. Agents are good at local progress: make this screen work, add this endpoint, patch this bug. Debt appears when those local wins stop adding up to one system people understand.
• Weeks 1-2 feel productive: the app exists and the backlog moves. By week 6, repeated prompts have created competing patterns and weak tests. By month 3, every change requires rediscovery: what the code does, why it does it, and which parts are safe to touch.
• The value of an AI technical debt audit is not a longer list of problems. It is a clearer map of what is making change expensive: which risks are urgent, which patterns are spreading, which tests do not prove the business flow, and which repo instructions need to catch up before the next agent run.

Audit Target and Version

• Repository: checkout-service
• Default branch: main
• Checked commit: mocked-build-reference

What We Checked

• Duplicated logic and near-identical modules
• Unstable abstractions and code paths that resist safe change
• Missing tests on critical paths and tests that do not prove real behavior
• Dependency drift and lockfile mismatches
• Secrets, unsafe config, or access-control assumptions left in reachable code
• Dead code, context rot, and modules nobody wants to touch
• Product intent missing from code, tests, or agent instructions

Findings