Tracing Technical Debt by AI-Based Correlation Agent : A Case Study of Nokia Corporation
Grönroos, Satu; Hietanen, Tuula; Salmi, Anni; Turtiainen, Katja (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-2025120231583
https://urn.fi/URN:NBN:fi:amk-2025120231583
Tiivistelmä
The objective of the project was to design and evaluate a solution that improves the efficiency of software maintenance processes and reduces technical debt within development environments. Nokia provided the broad development challenge on the theme of Artificial Intelligence, thereby serving as the commissioning organization for the project's overall scope. The project team then specified the purpose to create and assess a concept for an AI-powered Correlation Agent capable of identifying root causes of technical debt and supporting proactive maintenance.
The development task was to conceptualize and prototype an AI agent that integrates structural code data with dispersed textual contexts such as Git commit messages and documentation, in order to diagnose and mitigate technical debt. Nokia benefits from reduced maintenance costs, faster workflows, and improved code quality based on the potential of this solution.
The study builds on theories of technical debt management, agentic AI versus generative AI capabilities, and correlation analysis as a statistical method for identifying relationships between key indicators and maintenance goals. The research employed a Design Sprint approach complemented by brainstorming sessions, current-state mapping, benchmarking of existing tools (CodeScene, SonarQube, Greptile, Powerdrill), and structured questionnaire interviews with IT Systems Specialists.
The tools that are already available do not combine code analysis with contextual information, which makes it more difficult to solve problems. Agentic AI, due to its inherent multi-step reasoning capabilities, is highly suitable for multi-step maintenance workflows and proactive risk detection. A prototype named Code Colleague (CoCo) demonstrated that a conversational interface enhances collaboration between developers and the AI agent.
The Correlation AI Agent can significantly reduce manual investigation time, accelerate fault correction, and lower long-term maintenance costs. The focus of future development should be on integrating the Correlation AI Agent into a multi-agent chain, defining inter-agent collaboration protocols, and establishing auditing mechanisms for AI-driven processes.
Keywords: Technical Debt, Agentic AI, Correlation AI Agent, Software Maintenance, Design Sprint
The development task was to conceptualize and prototype an AI agent that integrates structural code data with dispersed textual contexts such as Git commit messages and documentation, in order to diagnose and mitigate technical debt. Nokia benefits from reduced maintenance costs, faster workflows, and improved code quality based on the potential of this solution.
The study builds on theories of technical debt management, agentic AI versus generative AI capabilities, and correlation analysis as a statistical method for identifying relationships between key indicators and maintenance goals. The research employed a Design Sprint approach complemented by brainstorming sessions, current-state mapping, benchmarking of existing tools (CodeScene, SonarQube, Greptile, Powerdrill), and structured questionnaire interviews with IT Systems Specialists.
The tools that are already available do not combine code analysis with contextual information, which makes it more difficult to solve problems. Agentic AI, due to its inherent multi-step reasoning capabilities, is highly suitable for multi-step maintenance workflows and proactive risk detection. A prototype named Code Colleague (CoCo) demonstrated that a conversational interface enhances collaboration between developers and the AI agent.
The Correlation AI Agent can significantly reduce manual investigation time, accelerate fault correction, and lower long-term maintenance costs. The focus of future development should be on integrating the Correlation AI Agent into a multi-agent chain, defining inter-agent collaboration protocols, and establishing auditing mechanisms for AI-driven processes.
Keywords: Technical Debt, Agentic AI, Correlation AI Agent, Software Maintenance, Design Sprint