Imagine relying on a tool designed to enhance your productivity, only to find it hindering your performance instead. This is the surprising conclusion drawn from a recent study by the AI research nonprofit METR.
Contrary to popular belief that AI assists seasoned software developers, the research found that experienced coders actually slowed down when using AI tools like Cursor on projects they were already familiar with. The study examined veteran developers and their expectations regarding AI’s impact on their work.
Many developers anticipated that AI would save them time, estimating a reduction in task completion time by nearly 25%. Surprisingly, even after using the AI, they perceived only a modest increase in speed.
In reality, the data revealed that AI increased the time required to complete tasks by 19%. Researchers Joel Becker and Nate Rush expressed their shock at these results, which challenge the assumption that AI can reliably enhance productivity for skilled engineers.
This study emerges alongside claims from tech leaders, suggesting that AI could displace many entry-level white-collar jobs in the near future. While previous studies touted significant productivity gains, METR’s findings suggest a different narrative—one where AI struggled to provide the expected benefits in familiar, complex codebases.
Participants in the study often found themselves needing to verify and correct AI suggestions, which were frequently along the right lines but not exact enough to trust without scrutiny. This resulted in more time spent revising and sometimes discarding AI output.
Interestingly, these insights may also reflect broader experiences with AI tools in everyday smartphone usage. Many users rely on AI features, such as predictive text and voice assistance, expecting them to simplify tasks.
However, these features often require careful review, making users question their convenience. In a technology-driven world where AI is prominent, it is essential to recognize both its potential and its limitations.