AI-Powered Coding Tools Boost Developer Productivity by Over 26% in Major Field Study

Boston, MA – A significant new study involving nearly 5,000 software developers across major companies has revealed that the use of AI-powered coding assistants, specifically GitHub Copilot, leads to a substantial 26.08% increase in completed tasks. The findings, detailed in a February 2025 paper titled "The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers," were highlighted by Wharton professor Ethan Mollick, a prominent voice on AI's impact on work.

The research, co-authored by Kevin Zheyuan Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng, and Tobias Salz, conducted randomized controlled trials at Microsoft, Accenture, and an anonymous Fortune 100 company. These real-world experiments tracked metrics such as pull requests, commits, and code builds, demonstrating significant productivity gains, particularly among less experienced developers who showed higher adoption rates and greater benefits from the tool. The version of Copilot used was based on GPT-3.5, an older iteration of the technology.

Mollick, known for his work on AI and the future of labor, acknowledged the study's findings in a tweet, stating:

"This clashes with some large-scale field experiments of coders working at companies, but worth noting."

This statement likely alludes to the nuanced impact of AI, a concept Mollick explores in his broader research on the "jagged technological frontier." While AI can dramatically enhance performance within its capabilities, its effectiveness can diminish or even become counterproductive when applied to tasks outside this frontier, requiring careful human judgment and oversight.

The study underscores the profound and immediate impact of generative AI on high-skilled work, even with earlier models. Experts suggest that such productivity increases necessitate a re-evaluation of workflows and a strategic approach to AI integration beyond mere cost-cutting. The findings further emphasize that organizations must foster a culture of experimentation and understanding to fully leverage AI's potential, ensuring that human expertise guides its application for optimal outcomes.