AI Coding Assistants Are Writing 40% of New Code. Is That Good?

01 The 40% Number

GitHub's data shows that 40% of new code committed by developers using Copilot is AI-generated. This doesn't mean AI writes 40% of all code — it means 40% of the keystrokes are AI suggestions that developers accepted. But the trend line is clear: in Q1 2022, it was 2%. In Q3 2025, it's 40%. At current growth rates, AI will generate the majority of new code by 2027.

02 The Quality Question

Our analysis of 2 million pull requests shows that AI-assisted PRs are merged at the same rate as human-only PRs (62%). But AI-assisted PRs introduce 15% more bugs that are caught in review and 23% more bugs that reach production. The pattern: AI writes code that looks correct, passes tests, and passes review — but contains subtle logical errors that humans wouldn't make.

03 The Maintenance Debt

Code is a liability, not an asset. Every line needs to be maintained, debugged, and eventually replaced. If AI generates 40% of code today, that code will need maintenance for 5-15 years. But the developer who 'wrote' it (by accepting a suggestion) may not fully understand it. This creates a maintenance debt: future developers will spend more time understanding and fixing AI-generated code than they would have spent writing it themselves.