#atom

When to refactor duplicated code in AI-assisted development

Core Idea: While traditional software engineering suggests refactoring after code is duplicated three times, LLMs naturally tend toward duplication unless explicitly instructed to refactor.

Key Elements

• Traditional "Rule of Three" suggests:
  • Duplicate code once (two copies) is acceptable
  • On third occurrence, refactor to eliminate duplication
  • Balances DRY (Don't Repeat Yourself) with premature abstraction risks
• LLMs have strong tendencies toward code duplication:
  • Models naturally produce complete copies with modifications
  • Require explicit prompting to perform refactoring
  • Will follow existing codebase patterns of duplication
  • May not proactively identify refactoring opportunities

Implementation Strategies

• Explicitly instruct LLMs to reduce duplication when generating code
• Ask models to identify and factor out common patterns
• Request the creation of helper methods when writing similar tests
• Be aware that in agent mode, models need to discover helpers to use them
• Review generated code specifically for duplication patterns

Balancing Concerns

• Excessive DRY can lead to wrong abstractions
• Too little refactoring creates maintenance challenges
• Context window limitations may prevent models from seeing all duplicated code
• Balance between specificity (duplication) and abstraction (reuse)
• Consider readability and maintainability in refactoring decisions

Connections

• Related Concepts: Preparatory Refactoring (code restructuring), Culture Eats Strategy (environment influence)
• Broader Context: Code Quality Principles (software design fundamentals)
• Applications: DRY Principle Implementation (reducing repetition)

References

1. Sandi Metz (2016). "The Wrong Abstraction." https://sandimetz.com/blog/2016/1/20/the-wrong-abstraction
2. Edward Z. Yang (2025). "AI Blindspots" collection, March 2025.

#refactoring #code-quality #duplication #abstraction

Connections:

Sources:

• From: AI Blindspots