Recent findings from the nonprofit METR challenge the effectiveness of AI coding assistants in boosting developer productivity. Despite assumptions that tools like GitHub Copilot and Cursor Pro enhance efficiency, the study revealed that experienced developers experienced a 19% slowdown when using AI tools, contrary to their expectations of a 24% time reduction. Key reasons for this decline include the need for extensive code review and rewrites due to reliability issues. Although AI tools show potential in creating new systems, they perform poorly in maintaining complex, existing codebases. Developers also encountered a learning curve, with only 56% familiar with Cursor. While some studies report productivity gains in specific contexts, METR’s findings highlight significant variability influenced by task complexity and tool familiarity. With ongoing improvements expected in AI coding tools, the software community remains divided on their long-term impact, emphasizing the necessity for human oversight in the coding process.
Source link
Share
Read more