Ever wondered what could be slowing down even the most experienced coders? Surprisingly, it’s AI. Ironically, the very technology created to enhance productivity is now becoming a bottleneck for seasoned developers.
A recent study challenges the popular belief that AI tools are game-changers for senior software engineers. Instead of boosting their speed, AI actually slowed them down when working on projects they already knew inside and out.
Conducted by METR, an AI research nonprofit, the study focused on veteran developers using Cursor, a popular AI coding assistant, to work on open-source projects they were familiar with. The developers initially thought AI would save them time, estimating that it could cut task completion by up to 25%. However, after using the tool, many still felt they were around 20% faster. But the data told a different story: AI actually increased the time needed to complete tasks by 19%.
Joel Becker and Nate Rush, who led the research, were surprised by the results. Rush, in particular, had expected AI to double productivity. This expectation is one that many of us share. However, the findings questioned the widely held assumption that AI can reliably boost the productivity of experienced software engineers-especially those working on large, complex projects that they know well.
The study revealed that instead of speeding up the process, AI tools required developers to spend additional time reviewing and correcting the AI’s suggestions. Though the AI often pointed them in the right direction, it rarely delivered exactly what was needed, leading to increased time spent reviewing, editing, and sometimes discarding the AI’s output.
While the study results may not apply universally, they highlight a significant issue. Developers who are already highly skilled with extensive experience working on known codebases did not benefit from AI in the way they had hoped. However, less experienced developers or those working on unfamiliar projects might still see a productivity boost from AI.
The study’s implications extend beyond the realm of software development. It gives us a glimpse into how the everyday smartphone user interacts with AI. From predictive text and voice assistants to photo editing, many of us have experienced AI features that seem more time-consuming than helpful. Whether it’s correcting an autocorrect mistake, fixing a poorly edited photo, or sorting through confusing AI-generated messages, we’re realizing that convenience sometimes comes at the cost of extra time.
Ultimately, this study serves as a reminder that AI is far from perfect, and its impact on productivity is far more nuanced than we might expect.