AI Coding Assistants: Boosting Productivity While Challenging Code Quality

2023 marked a breakthrough year for GitHub Copilot, which in less than two years evolved from a prototype to an essential tool for many developers and enterprises, heralding a new era of AI‑assisted programming.

GitHub’s own research reports a 55% increase in coding speed when using Copilot, but raises the question of how the quality and maintainability of AI‑generated code compare with manually written code.

GitClear collected 153 million code‑change records from January 2020 to December 2023, creating the largest known structured dataset for analysing code‑quality differences.

The analysis reveals worrying trends: the code‑change rate (the proportion of lines modified or reverted within two weeks) is projected to be twice the 2021 level by 2024. The share of “added” and “copy/paste” code is rising, making AI‑generated code resemble the work of a frequently changing contract worker and often violating the DRY principle.

Based on these findings, the paper offers several managerial recommendations for maintaining high‑quality code in an AI‑driven environment.

GitHub headline: “AI coding improves efficiency by 55%, adds 46% more code, and contributes $1.5 trillion to GDP.”

GitHub CEO Thomas Dohmke authored a 2023 blog post detailing Copilot’s rapid adoption and its economic impact.

More than 20 000 organizations now use Copilot for Business, and over one million developers have adopted the personal version. A June 2023 study with Wakefield Research found that 92% of developers at large U.S. companies use AI coding tools, with 70% reporting significant benefits, while an August 2023 O’Reilly survey showed that 67% of respondents had not yet used ChatGPT or Copilot, indicating further market potential.

Developers embrace Copilot for its speed gains, yet senior engineers worry about maintenance. Adam Tornhill (author of *Your Code as a Crime Scene*) and Robert Martin (author of *Clean Code*) highlight that faster code writing can lead to higher future reading costs and lower overall quality.

Key challenges identified include:

Frequent suggestions to add code but few to update, move, or delete it, stemming from UI constraints.

Time‑consuming evaluation of multiple competing suggestions, especially in IDEs like JetBrains.

Suggestion algorithms aim for acceptance, not for reducing maintenance effort, creating a misalignment of incentives.

These issues partly explain why junior developers accept AI suggestions about 20% more often than senior developers.

The study classifies code changes into seven categories (added, deleted, moved, updated, find/replace, copy/paste, and invalid operations) and analyses their yearly distribution from 2020 to 2024, showing increases in additions, deletions, updates, and copy/paste, while moves decline sharply.

Trend analysis using OpenAI’s gpt‑4‑1106‑preview model predicts continued growth of these patterns, with notable signals such as a 39.2% rise in code churn, a 17.3% drop in moves, and an 11.3% rise in copy/paste.

The surge in churn suggests more code is being discarded shortly after being merged, raising the risk of buggy code reaching production. If this continues, the proportion of changes reverted within two weeks could exceed 7%, double the 2021 level, potentially impacting DevOps metrics like change‑failure rates.

Reduced code movement hampers refactoring and reuse, while increased copy/paste inflates technical debt and future maintenance difficulty.

Open research questions include: how to incentivize reduction of “add‑then‑forget” patterns, the exact impact of extra code on development speed, and whether copy/paste will reach 20‑25% of all code operations by 2024.

In conclusion, the data indicate a clear decline in code quality in 2023 linked to the widespread use of large‑language‑model‑based assistants. Developers now prioritize code quality and production incident metrics, highlighting the need for tools and processes that balance AI‑driven productivity with long‑term maintainability.