Research project · March–October 2026
AI coding assistance and maintainability in JetBrains repositories
An observational study of 199 JetBrains open-source repositories, examining AI coding assistance through six maintainability proxies.
GitHub repositorySampling frame documented in the public repository.
Research question
Is AI-assisted coding associated with changes in code maintainability? This repository-mining study examines open-source projects from JetBrains and JetBrains-Research. A reproducible random sample selects 199 repositories from 411 eligible code repositories, drawn from a population of 1,049.
Study design
The date-based analysis compares 2019–June 2021 with 2023–2025 using comment density, commit size, bug-fix ratio, cyclomatic complexity, function length and deletion/addition ratio. The methodology specifies Mann–Whitney U tests, effect sizes and Benjamini–Hochberg correction for multiple comparisons. A second analysis compares explicitly AI-labelled and unlabelled commits within the same repository and months.
Current evidence
The public authorship summary records 1,277 AI-labelled commits across 24 of the sampled repositories. These labels document declared assistance, not all AI use. Code, sampling records, analysis tables and methodology are available on GitHub; there is no final report. The repository documents corrections to comment counting and language coverage, so earlier charts should not be treated as final results.
Interpretation and limitations
Neither design establishes a causal effect of AI assistance. Changes over time may reflect project maturity or team composition; developers also choose which tasks receive assistance. Unlabelled commits may still be AI-assisted, and maintainability proxies are not direct measurements of long-term software quality. Generalisation beyond this ecosystem requires further evidence.
Related research
GPTSwarm: evolutionary graph optimisation