Study Reveals AI Coding Tools May Slow Down Experienced Developers
New research indicates AI coding tools may slow down experienced software developers.
Key Points
- • AI tools can slow down experienced developers, contrary to expectations of improved speed.
- • The METR study shows task completion times increased by 19% rather than decreased.
- • Developers often spend time correcting AI suggestions, detracting from efficiency.
- • Despite the slowdown, many developers find value in using AI tools for improved project enjoyment.
A recent study by the AI research nonprofit METR, published on July 13, 2025, has found that the use of AI coding tools can actually hinder productivity among experienced software developers. The study focused on seasoned developers leveraging Cursor, a popular AI coding assistant, for their tasks in open-source projects. Surprisingly, although developers initially expected AI to reduce their task completion time by 24%, they ended up experiencing a 19% increase in completion time instead.
The slowdown has been attributed to the necessity for developers to spend additional time correcting AI suggestions, which, while often directionally correct, were not precisely applicable to the task at hand. The lead authors of the study, Joel Becker and Nate Rush, expressed their astonishment at the findings, as Rush had anticipated tangible gains in speed from using the AI tool.
This research emerges in contrast to prior studies that have reported significant productivity improvements afforded by AI, with some reporting speed increases as high as 56% in coding tasks. The findings suggest that these benefits are not universal and may vary greatly depending on the developer's experience and the complexity of the codebase. Becker noted, "The AI's recommendations were often directionally correct but not exactly what was needed," leading to inefficiencies among skilled developers who were already familiar with the work.
While the implications of this study could affect the interest in AI products for software development, the authors caution that the slowdown noted may not hold true for junior engineers or developers unfamiliar with certain codebases.
Despite the slowdown, Becker highlighted that many developers continued enjoying the use of Cursor, indicating that these AI tools might enhance the overall development experience beyond just acceleration. He stated, "Developers have goals other than completing the task as soon as possible," suggesting that the usability and experience improvements provided by AI tools may still hold value for seasoned professionals in tech.