2025 research on AI code quality:

linkGoogle Doc

https://docs.google.com/document/d/1KFcI0M-UHTAGdSstqZ4XfHKYM80XcSERfTEl1tXPkj0/edit?tab=t.0

linkFinal Version

Some possible ways to frame GitClear's 2025 code quality research:

1. Coding very fast...towards what? Would you hire a Junior developer and let them copy/paste thousands of lines in your repo? If not, you might want to pay attention to what's creeping into your repo.

2. Wanted: Conscientious developers. In an environment where AI is empowering even junior devs to become prolific contributors, it has never been more valuable to differentiate on "attention to details." If you don't, research shows your AI assistant will begin generating churn and defects.

3. One-sided AI incentives. Over the past year, there has seemingly been a new story every week about an AI breakthrough that accelerates a developer's frequency of authorship. What is conspicuously absent? Research like GitClear's that question whether there will be long-term costs to today's AI feeding frenzy. The companies funding AI have zero incentive to articulate the risks, if they even understand them.

4. The year that copy/paste won. According to GitClear's latest research, utilizing the largest known set of structured code change data (211 million changed lines from Microsoft, Google, etc over 5 years), this was the first year that "Copy/paste" beat "Moved" code in prevalence. Meanwhile, Google reports that every 25% increase in AI adoption corresponds with an 8% increase in defects.

5. The race toward code-refactoring AI. GitClear's recent research proves the risk that companies assume when they trust their development to AI assistants that only add code. What's in the works to shift the balance of new code back toward the pre-AI norms, and will developers use it?