A study shows that AI-enhanced development tools do not necessarily increase productivity in software development. Conducted with experienced open-source developers, the researchers found that those using AI took 19% longer to complete tasks than those without AI. This was contrary to the expectation of a 40% speedup. Participants worked on realistic issues from large open-source projects, and both objective and subjective metrics were collected. The findings expose a disparity between the perceived efficiency of AI tools and their real-world effects on developer performance.
The study revealed that developers using AI tools, including Claude 3.5 and Cursor Pro, experienced a 19% increase in task completion time despite believing they were working faster.
Researchers conducted their study using experienced open-source developers in production-grade environments, rather than synthetic benchmarks, to assess the real effects of AI on software development.
Eyes were on a gap between AI's perceived benefits and its actual impact, as developers and external experts expected a productivity increase of ~40%, but results showed a significant slowdown instead.
Only 16 professional developers with an average of five years of experience participated, tackling realistic issues from large, mature open-source projects across 246 tasks.
#ai-in-software-development #productivity-challenges #open-source-development #software-engineering #developer-experience
Collection
[
|
...
]