AI coding tools have delivered on raw output, but the cleanup is showing up in review queues and maintenance work. BairesDev’s Q3 2026 Dev Barometer, fielded in August among 705 developers in more than 60 countries, found 42% now say AI writes at least half their code, up from 12% a year earlier, WebProNews reported. Sixty-seven percent said they spend more time reviewing AI-generated code than a year ago, and 52% said they spend more time debugging problems the models introduced.
Issues that stick around
A large study posted to arXiv, “Debt Behind the AI Boom,” examined about 302,600 verified AI-authored commits across 6,299 GitHub repositories, covering tools including Copilot, Claude, Cursor, Gemini and Devin. The researchers identified 484,366 AI-introduced issues, of which 89.3% were code smells, with the rest split between correctness and security problems. Between 7.5% and 14.8% of commits introduced at least one issue, depending on the tool, and 22.7% of the tracked issues were still unresolved in the repositories’ latest revisions.
More pasting, less refactoring
GitClear’s 2026 code quality research, based on 623 million analyzed code changes, found duplicated code blocks rose 81%, from 40.3 to 73.0 per million changed lines between 2023 and this year to date. Moved code, which GitClear uses as a stand-in for refactoring, fell from 21% of changed lines in 2022 to 3.8%, while copy-pasted lines rose from 9.4% to 15.7%. The pattern suggests teams are adding new code faster than they consolidate what already exists.
Quality is a process problem
Not everyone reads the data as proof that AI code is inherently worse. In a widely shared essay titled “If AI coding is lowering your code quality, you’re not managing quality right,” the author argues that quality suffers mainly when teams “blindly merge the PRs and send them off to prod.” With a layered approach, the author writes, it’s possible to keep bug counts stable or even lower them while increasing output.
The setup described includes spec-driven development, where an AI reviews requirements or a technical design for gaps and edge cases before any code is written, and test-driven development done carefully, so an agent doesn’t simply write tests that pass the bugs it just introduced. The author credits the requirements-review step with a sharp drop in bugs in freshly written code.
The BairesDev survey points the same way from the budget side: 78% of the 41 enterprise CTOs surveyed said they had increased spending on code review, quality assurance or validation to support AI-generated work.


WebProNews is an iEntry Publication