AI Coding Tools Increase Technical Debt and Slow Development

Muzammil Rawjani, founder of Software Orca, argues that the perceived productivity gains from AI coding tools mask a decline in software quality and long-term maintainability. While Gartner reports that 90% of engineering leaders see productivity improvements averaging 19.3%, other data suggests a different reality.

A randomized controlled trial by METR found that AI-assisted developers were 19% slower than those without AI, despite the developers predicting a 24% improvement in their own performance. This gap between sentiment and reality is compounded by a decline in system simplification.

Data from Git Clear shows that code refactoring dropped from 25% to under 10% of all changes between 2021 and 2024 as AI tools became mainstream. Rawjani suggests this shift indicates that developers are prioritizing rapid code generation over maintainability, which increases technical debt.

To address these issues, Rawjani recommends that organizations implement strict review gates for AI-generated code. He further advises tracking concrete delivery metrics and defect rates rather than relying on developer sentiment to measure productivity.