Generatiivinen tekoäly ja ohjelmistokehittäjien työn tuottavuus – Pilottitutkimus suomalaisessa teknologiakonsulttiyrityksessä
Year: 2025
Type: dissertation
Study designs: Randomized Controlled Trial, Clinical Trial
Abstract: This thesis examines the causal impact of generative AI tools on software developer productivity through a pilot randomized field experiment. As organizations rapidly adopt these technologies despite limited causal evidence, understanding their productivity effects becomes critical for informed investment decisions. Using a three-arm randomized controlled trial with 25 professional developers across 130 task observations at a Finnish technology consultancy, I study whether access to generative AI tools affects development speed and code quality, and whether structured training on using the tools moderates these effects. Developers given GenAI access finish standardized tasks 40.2 percent faster and write 26.2 percent higher quality code according to an expert judge, though wide confidence intervals reflect limited statistical precision from the small sample. Lee (2009) bounds suggest these effects persist under extreme selection assumptions, despite high differential attrition between the treatment groups. Additionally, structured training appears to amplify quality improvements, whereas time savings emerge even from access alone, hinting at complementarities between technology and human-capital investments. This pilot study demonstrates the feasibility of randomized field experiments for evaluating workplace AI tools while highlighting critical methodological challenges. The 35.9 percent overall attrition rate and post-hoc power of only 25 percent for time effects and 19 percent for quality improvements underscore the need for larger samples in future research. Results provide suggestive evidence that generative AI tools may enhance developer productivity, particularly when coupled with appropriate training and implementation support.
Source: Aaltodoc (Aalto University)
Author: Häkkinen, Petteri
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7019479082