GitClear says AI coding users gained 25 percent output since 2023, and block duplication in their code rose 81 percent over the same period

GitClear's Maintainability Gap report, covering 623 million analyzed code changes from 2023 through the first half of 2026, says heavy users of AI coding tools produce 25 percent more code than they did before, while block duplication in their commits has risen 81 percent and refactoring activity has collapsed. The report was published in June 2026 by GitClear, a code review company that sells analysis of these signals. The New Stack's Steve Fenton wrote up the findings on 14 September and quoted the specific figures.

What the numbers say

GitClear measures block duplication per million changed lines. That number climbed from 40.3 in 2023 to 73.0 year-to-date in 2026, an 81 percent rise and, according to GitClear, the highest level on record. Moved code, which the report treats as the fingerprint of refactoring, fell from 21 percent of changed lines in 2022 to 3.8 percent in 2026. Within-commit copy and paste is up 41 percent, error-masking constructs are up 47 percent, and two-week code churn is up 15 percent.

GitClear frames the output gain plainly. Heavy AI users gained 25 percent on their own prior velocity. Vendor marketing has circulated 10x figures; GitClear's dataset does not show that. The report also finds those same heavy users producing 4 to 10 times more output than non-AI peers, and Fenton adds that the same teams were already ahead of the industry before AI tools arrived. Fast teams stayed fast; slow teams did not close the gap.

Before AI, GitClear reported a 2x preference for refactoring over redundant copy and paste. In the first half of 2026, developers are 5x more likely to paste than to refactor. That ratio flipped in 2024, which the report calls the first year on record where within-commit copy and paste exceeded moved code.

The source and its stake

GitClear sells a product that grades these signals, so it has a commercial interest in the finding. The dataset is real and the sample is large, but every conclusion in the report has to be read as GitClear's own. Both charts in the New Stack piece are labelled to GitClear, and the underlying report is available on the company's site behind a contact form.

A rise in block duplication is what a code review discipline is designed to catch, and what an agent working alone will let through. The report reads as a description of what happens when an AI writes at pace and nothing reads the diff for structural quality before it lands. Fenton's own conclusion is the same: the teams doing well with AI are the teams that were doing well before AI, because they maintain the technical practices the tools do not replace.

A team paying for Claude Code, Cursor or Copilot can use the report as an external number in a budget conversation. The 10x figure has no support in GitClear's 623 million changes, and the same tools that deliver a 25 percent gain also raise the maintenance bill. Any team adopting these tools should budget review time to remove duplication before drift across copies of the same block produces repeat bugs, and should measure their own signals rather than assume the vendor benchmark applies to their codebase.

Source

The New Stack: Your AI coding spend bought 25 percent more output. Duplication rose 81 percent., citing GitClear: The Maintainability Gap, AI Code Quality in 2026.