How much does AI impact development speed? An enterprise-based randomized controlled trial
Authors:
Elise Paradis,
Kate Grey,
Quinn Madison,
Daye Nam,
Andrew Macvean,
Vahid Meimand,
Nan Zhang,
Ben Ferrari-Church,
Satish Chandra
Abstract:
How much does AI assistance impact developer productivity? To date, the software engineering literature has provided a range of answers, targeting a diversity of outcomes: from perceived productivity to speed on task and developer throughput. Our randomized controlled trial with 96 full-time Google software engineers contributes to this literature by sharing an estimate of the impact of three AI f…
▽ More
How much does AI assistance impact developer productivity? To date, the software engineering literature has provided a range of answers, targeting a diversity of outcomes: from perceived productivity to speed on task and developer throughput. Our randomized controlled trial with 96 full-time Google software engineers contributes to this literature by sharing an estimate of the impact of three AI features on the time developers spent on a complex, enterprise-grade task. We found that AI significantly shortened the time developers spent on task. Our best estimate of the size of this effect, controlling for factors known to influence developer time on task, stands at about 21\%, although our confidence interval is large. We also found an interesting effect whereby developers who spend more hours on code-related activities per day were faster with AI. Product and future research considerations are discussed. In particular, we invite further research that explores the impact of AI at the ecosystem level and across multiple suites of AI-enhanced tools, since we cannot assume that the effect size obtained in our lab study will necessarily apply more broadly, or that the effect of AI found using internal Google tooling in the summer of 2024 will translate across tools and over time.
△ Less
Submitted 11 November, 2024; v1 submitted 16 October, 2024;
originally announced October 2024.
