Central Theme
The article examines whether AI coding assistants actually improve developer productivity, reporting on a study that challenges the popular narrative of AI-driven efficiency gains in software development.
Key Findings
- A controlled study by research group METR found that experienced developers using AI tools were 19% slower at completing real-world tasks (bug fixes, new features) in large, open-source projects.
- A significant perception gap was observed: despite being slower, developers believed the AI tools made them 20% faster.
- The study identifies five likely reasons for the slowdown: developer over-optimism, high familiarity with the codebase (making AI less useful), the complexity of large repositories, low reliability of AI suggestions (less than 44% were accepted), and the AI’s lack of implicit contextual understanding.
- Developers using AI spent less time coding and more time prompting, waiting for, and reviewing AI outputs, which added significant overhead.
Conclusion
In the context of experienced developers working on large, familiar codebases, current AI tools can decrease productivity rather than enhance it. The time spent managing and verifying AI output can outweigh the benefits of code generation. This suggests that the utility of AI coding tools is highly situational and that perceived productivity gains should be critically evaluated with objective data.
Mentoring Question
Considering the study’s findings on the gap between perceived and actual productivity, how can you and your team objectively measure the impact of a new tool to ensure its adoption is based on real efficiency gains, not just hype?
Source: https://www.theregister.com/2025/07/11/ai_code_tools_slow_down/