AI Coding Tools Boost Velocity by 25% but Spike Code Duplication by 81%
A massive analysis of 623 million code changes reveals a decline in refactoring and a rise in unsustainable copy-pasting.
The promise of a "10x developer" powered by artificial intelligence is colliding with a stark reality of declining code quality. While AI tools have marginally increased the speed of software production, they are simultaneously eroding the structural integrity of the codebases they help create.
According to a report by GitClear, which analyzed 623 million code changes between 2023 and 2026, heavy AI users saw a 25% increase in velocity compared to their own prior performance. While this represents a measurable gain, it falls far short of the exponential productivity leaps often claimed by AI vendors. More concerning is the cost of this speed: block duplication rose by 81% over the same period, climbing from 40.3 to 73.0 instances per million changed lines.
The Collapse of Refactoring
The data indicates a fundamental shift in how developers interact with their code, marking a return to a "code-and-fix" era. Refactoring—the process of restructuring existing code to improve its internal quality without changing its external behavior—has plummeted. GitClear found that code refactoring, measured as "moved code," fell from 21% of changed lines in 2022 to just 3.8% by 2026.
This decline has flipped the traditional developer workflow on its head. Before the widespread adoption of AI, developers were twice as likely to refactor as they were to copy and paste. Now, the ratio has reversed; developers are roughly five times more likely to copy and paste code than to refactor it.
The Maintainability Gap
This trend creates what is known as a "maintainability gap." By prioritizing immediate output over structural health, AI tools are encouraging the creation of tightly coupled and redundant codebases. While the short-term result is a faster delivery of features, the long-term consequence is a codebase that becomes increasingly incomprehensible and expensive to maintain.
As duplication rises and refactoring vanishes, the risk of "whack-a-mole" bugs increases—where fixing a flaw in one section of the code leaves identical flaws untouched in duplicated blocks elsewhere. Over time, this technical debt can become so burdensome that it stalls future feature development entirely, as engineers spend more time navigating redundancy than writing new logic.
What to Watch
As the initial hype surrounding AI productivity begins to cool, the industry must now determine if a 25% increase in velocity is worth the systemic risk to code quality. The GitClear findings suggest that the perceived gains in speed may be an illusion, masking a long-term decline in software sustainability. The critical question remaining is whether AI tools can be evolved to prioritize refactoring and architectural health, or if they will continue to incentivize a culture of disposable, duplicated code.