When Clean Code Becomes Costly: Balancing Human Readability and AI Digestibility with the NUDEE Framework

Abstract

Aim/Purpose: This study maps the structural tension between human-centric Clean Code conventions and the architectural demands of LLM-based coding agents. It proposes NUDEE, a management framework for governing codebases optimized for both human readability and AI digestibility. Background: Clean Code prescribes deep abstraction and module decomposition to reduce human memory load. By 2026, LLMs and autonomous agents actively participate in coding workflows, but their token-limited context windows are penalized by the same decomposition that helps human developers. Methodology: A PRISMA 2020-guided systematic literature review of three databases (Scopus, Web of Science, ACM Digital Library) used a three-pillar Boolean search strategy followed by four-stage screening. The resulting corpus of 89 records (1996–2026) was inductively coded into five thematic streams. Contribution: The paper presents a PRISMA-structured systematic literature review – to our knowledge, one of the first to systematically map the human-readability–AI digestibility tension in software engineering – documents the field’s organizational turn (44.9% of the literature addresses management frameworks) and introduces the NUDEE framework. Findings: The dominant stream, hybrid and agentic SE management frameworks (44.9%), confirms that the field has reframed the problem as organizational; synthesis across all streams reveals three structural pillars for AI-optimal codebases. Recommendations for Practitioners: Development teams should adopt the Software Coach and System Orchestrator roles, instrument the Context Efficiency Index in CI/CD pipelines to detect semantic debt early, and revise code review criteria to include digestibility alongside correctness. Recommendation for Researchers: Empirical validation of the CEI against real agent task-solve rates in production codebases is the field’s most pressing research need. Impact on Society: As AI agents increasingly execute code in critical systems, semantic debt in organizational codebases directly translates into reliability and safety risk. NUDEE provides a governance vocabulary that enables mid-size organizations, not only large technology firms, to manage human AI software knowledge assets responsibly. Future Research: CEI empirical validation with threshold calibration and longitudinal case studies testing whether teams adopting the dual-role model achieve higher agent reliability and shorter release cycles.