How Effective are AI-powered Code Assistants in Enhancing Developer Productivity?
Thaw, Thin Thu Thu (2025)
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2025061122253
https://urn.fi/URN:NBN:fi:amk-2025061122253
Tiivistelmä
The introduction of artificial intelligence (AI) technologies into software development has transformed pro
gramming activities. AI-based coding assistants, such as GitHub Copilot, Tabnine, and Codeium, support de
velopers by generating code and providing contextual recommendations in real time. This research anal
yses how these tools impact productivity for developers with different levels of skills and varying degrees of
task complexity.
A mixed-methods design was adopted wherein quantitative data was collected from 100 developers
through structured surveys, and qualitative data was gathered through 10 semi-structured interviews.
The thesis studied the effect of four independent variables on the dependent variable of developer produc
tivity: the quality of code suggestions, task complexity, tool user-friendliness, and the user’s level of exper
tise. The mean scores of all constructs suggested strong, positive perceptions concerning the usefulness of
AI tools, as captured through descriptive statistics.
Correlation analysis indicated all independent variables to have significant relationships with productivity;
in particular, the quality of code suggestions had the highest correlation (r = .798). Subsequent analysis via
multiple regression confirmed only the quality of suggestion to be a significant predictor of productivity (β =
1.162, p < .001). Suggestion usability, suggesting expert level, and task complexity while being controlled
for each other exhibited varying or no significant impacts.
Thematic analysis pointed to user trust, efficiency, and learning as primary advantages, while focus on over
dependence, security, and ethics of AI coding raised other concerns. The results substantiated that AI
driven code assistants boost productivity when suggestions are precise and the integration of tools is seam
less. In addition, thoughtful design and monitoring are necessary to mitigate addiction and sustain software
standards.
gramming activities. AI-based coding assistants, such as GitHub Copilot, Tabnine, and Codeium, support de
velopers by generating code and providing contextual recommendations in real time. This research anal
yses how these tools impact productivity for developers with different levels of skills and varying degrees of
task complexity.
A mixed-methods design was adopted wherein quantitative data was collected from 100 developers
through structured surveys, and qualitative data was gathered through 10 semi-structured interviews.
The thesis studied the effect of four independent variables on the dependent variable of developer produc
tivity: the quality of code suggestions, task complexity, tool user-friendliness, and the user’s level of exper
tise. The mean scores of all constructs suggested strong, positive perceptions concerning the usefulness of
AI tools, as captured through descriptive statistics.
Correlation analysis indicated all independent variables to have significant relationships with productivity;
in particular, the quality of code suggestions had the highest correlation (r = .798). Subsequent analysis via
multiple regression confirmed only the quality of suggestion to be a significant predictor of productivity (β =
1.162, p < .001). Suggestion usability, suggesting expert level, and task complexity while being controlled
for each other exhibited varying or no significant impacts.
Thematic analysis pointed to user trust, efficiency, and learning as primary advantages, while focus on over
dependence, security, and ethics of AI coding raised other concerns. The results substantiated that AI
driven code assistants boost productivity when suggestions are precise and the integration of tools is seam
less. In addition, thoughtful design and monitoring are necessary to mitigate addiction and sustain software
standards.