AI-Assisted Defect Detection
Introduction
AI (LLMs like GPT-4, Claude, GitHub Copilot) provides semantic analysis to detect defects that traditional tools miss. Most valuable at Stages 1-3 (shift-left) and Stage 9 (automated approval).
AI Capabilities by CD Stage
| Stage | AI Capability | Defect Category | Value | How AI Helps |
|---|---|---|---|---|
| 1 | Ambiguous requirements detection | Knowledge & Communication | High | Flags vague terms, generates test scenarios, finds contradictions in specs |
| 1 | Test scenario generation | Testing & Observability | High | Generates Given/When/Then from requirements, exposes edge cases |
| 2 | Implicit knowledge detection | Knowledge & Communication | High | Flags magic numbers, undocumented business rules, missing "why" comments |
| 2 | Comprehensive test generation | Testing & Observability | High | Generates edge cases, negative inputs, boundary values, error paths |
| 3 | Semantic code review | Change & Complexity | High | Detects logic errors, missing validations, design smells beyond syntax |
| 3 | Divergent mental models | Knowledge & Communication | High | Finds terminology mismatches across services, bounded context violations |
| 5 | Integration test generation | Integration & Boundaries | Medium | Generates API test scenarios from contracts, cross-service test cases |
| 9 | Automated risk scoring | Process & Deployment | High | Analyzes change diff + history to auto-approve low-risk changes |
| 11 | Anomaly detection | Testing & Observability | Medium | ML-based alert thresholds, SLO burn rate prediction, pattern detection |
Next Steps
- Understanding defect categories? See External Defect Catalog
- Stage-by-stage prevention? Read Stage-by-Stage Guide
- Quality gates overview? Review Quality Gates Overview
- Using AI agents as contributors? See Agentic CD and ACD and the CD Model.
Tutorials | How-to Guides | Explanation | Reference
You are here: Explanation — understanding-oriented discussion that clarifies concepts.