AI code editors are increasingly common in software development, used by leaders like Amazon, Microsoft, and Google. However, a recent METR study reveals that experienced developers may actually slow down with AI tools, taking 19% longer to complete tasks compared to their non-AI peers. While AI-assisted developers believed their productivity increased by 20%, they spent significantly more time waiting for AI, prompting it, and reviewing its outputs. The study involved 16 seasoned developers, analyzing their performance with and without AI assistance.
Despite initial optimism that AI would speed up coding, the findings highlight major limitations of generative AI in software development. Developers accepted only 44% of AI-generated code and spent time correcting it. As AI tools evolve, the study suggests that the productivity impact varies by individual, urging developers to be judicious in their use of AI. This nuanced understanding could shape future adoption and integration of AI in coding practices.
Source link