Quantifying the impact of large Language Model (LLM) copilots on code cyclomatic complexity and developer velocity

Анотація

This paper investigates the impact of Large Language Models (LLM) integrated into development environments, exemplified by GitHub Copilot, on software code quality. The study focuses on the “Speed vs Maintainability” dilemma and the “productivity paradox” phenomenon, where the growth of volumetric performance metrics is accompanied by a decline in the code`s information density. To verify the “structural inflation” hypothesis, a controlled comparative experiment was conducted involving 40 mid-level developers divided into control and experimental groups. Developer velocity (V) and McCabe`s cyclomatic complexity (v(G)), quantified via static analysis (SAST) within a Python environment, were utilized as key metrics. The research confirms the existence of a statistically significant positive correlation between the utilization of generative AI assistants and an increase in code cyclomatic complexity. It is established that LLM, acting as stochastic optimizers of local context, tend to generate algorithmically redundant structures (“inflationary code”), leading to the exponential accumulation of hidden technical debt. It is concluded that code review protocols must be revised towards deep semantic audit and the implementation of “Klymenko`s thresholds” - rigid metric gateways developed to automatically reject high-entropy AI-generated structures.

Ключові слова: generative AI; LLM; developer velocity; github Copilot; human-inthe-loop

Як цитувати

@article{internauka12064,
  author   = {Klymenko, Serhii},
  title    = {Quantifying the impact of large Language Model (LLM) copilots on code cyclomatic complexity and developer velocity},
  journal  = {Міжнародний науковий журнал «Інтернаука»},
  year     = {2026},
  number   = {5},
  doi      = {10.25313/2520-2057-2026-5-12064},
  url      = {https://www.inter-nauka.com/issues/2026/5/12064}
}

TY  - JOUR
AU  - Klymenko, Serhii
TI  - Quantifying the impact of large Language Model (LLM) copilots on code cyclomatic complexity and developer velocity
JO  - Міжнародний науковий журнал «Інтернаука»
PY  - 2026
IS  - 5
SN  - 2520-2057
DO  - 10.25313/2520-2057-2026-5-12064
UR  - https://www.inter-nauka.com/issues/2026/5/12064
LA  - en
ER  -