AI has rapidly transformed the programming landscape, with claims that it generates 25% of Google’s code. However, a recent study from the nonprofit Model Evaluation and Threat Research (METR) reveals that AI tools may hinder, rather than help, software development. In a sample of 16 programmers tackling 250 coding tasks, those using AI tools like Anthropic’s Claude and Cursor Pro spent 19% more time completing tasks than their non-AI counterparts. Although AI tools reduced the time spent coding and debugging, they led to increased time spent reviewing outputs and fixing errors. Surprisingly, programmers believed AI would enhance their efficiency by 20%, despite the findings suggesting otherwise. Furthermore, “vibe coding” by untrained programmers raises cybersecurity risks, resulting in less secure code. As companies replace employees with automation, the quality and security of generated code have declined. This disconnect between expectations and reality may challenge the narrative of AI’s dominance in coding.
Source link
