AI slows down some experienced software developers, study finds

Introduction

A recent study has found that artificial intelligence coding assistants, such as GitHub Copilot, may actually slow down the productivity of experienced software developers, despite the growing popularity and ongoing investment in AI-powered developer tools. The findings challenge the prevailing narrative that AI will universally amplify productivity and ease for all programmers.

Study Overview

The research, led by Stanford University, was conducted in partnership with GitHub. The study tasked over 20 professional developers, each with an average of nine years' experience, with programming assignments using Copilot and without AI assistance. Contrary to earlier research with students and less experienced coders, this study revealed a surprising trend among professionals.

Key Findings:

  • Expert developers often slowed down when using AI coding assistants.
  • Experienced participants spent more time verifying—and frequently correcting—AI-generated code, which negated time savings.
  • Less experienced developers, in previous studies, found significant productivity boosts with tools like Copilot.

Why Are Experts Slower With AI?

While AI tools like Copilot suggest code snippets and automate boilerplate, advanced developers tended to distrust or double-check AI suggestions, leading to longer overall completion times. This increased authentication workload may explain the drop in productivity among seasoned professionals.

Verification Overhead

Skilled developers possess deep domain expertise and often write nuanced, complex code. They are accustomed to reviewing their own work critically; adding AI-generated material introduces another layer of review and potential debugging, which can slow down workflow.

AI Tools and the Future of Software Development

The study’s results highlight that while AI assistants like Copilot or ChatGPT are valuable for newer or less experienced programmers, their efficacy for veteran professionals is less clear-cut. The findings offer a nuanced perspective on how AI will fit into the workspace and what types of developers stand to benefit the most.

For a more detailed look at the research, see the original article at Reuters.

The study’s results highlight that while AI assistants like Copilot or ChatGPT are valuable for newer or less experienced programmers, their efficacy for veteran professionals is less clear-cut. The findings offer a nuanced perspective on how AI will fit into the workspace and what types of developers stand to benefit the most. The study’s results highlight that while AI assistants like Copilot or ChatGPT are valuable for newer or less experienced programmers, their efficacy for veteran professionals is less clear-cut. The findings offer a nuanced perspective on how AI will fit into the workspace and what types of developers stand to benefit the most.

Latest AI News

Stay Informed with the Latest news and trends in AI