Framework for sustainable integration of artificial intelligence in to software development life cycle : insights from state of art and state of practice
Mohamed Iqbal, Fathima Fazla (2025)
Diplomityö
2025
School of Engineering Science, Tietotekniikka
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe20251209116567
https://urn.fi/URN:NBN:fi-fe20251209116567
Tiivistelmä
Using Artificial Intelligence (AI) in the Software Development Life Cycle (SDLC) has significantly increased productivity and code quality. However, the sustainability implications of this transformation remain insufficiently understood across environmental, economic, social, individual, and technical dimensions.
Current research examines either technical benefits in isolated SDLC phases or the environmental footprint of AI systems in general. A holistic mapping of both positive (hand-prints) and negative (footprints) impacts across all phases and all five sustainability dimensions is missing, as is empirical evidence on real-world practitioner experiences and governance practices.
This thesis maps the sustainability impacts of AI in the SDLC and proposes a practical framework for its responsible integration. A mixed-methods approach was employed: 64 peer-reviewed studies (2020–2025) were used for a systematic literature review, and semi-structured interviews with 27 experienced software professionals from 12 countries representing diverse roles (junior to principal architects) and domains (financial services, telecommunications, aviation).
Key findings are: (1) AI use is heavily concentrated in the implementation and testing phases, with strong handprints in developer productivity, code quality, and testing efficiency. Still, it remains less in requirements engineering and maintenance despite academic recommendations. (2) Severe footprints are unmonitored energy consumption and carbon emissions, bias propagation, introduction of AI-specific bugs, and risk of developer skill degradation. (3) Practitioners prioritize immediate productivity gains and technical risks (security, code quality, technical debt) while remaining significantly less aware of environmental impacts than the academic literature suggests.
The results demonstrate that current AI adoption in software development achieves only “accidental” efficiency. This thesis outcome is a framework that consists of actionable components: mandatory training, human-in-the-loop oversight, and company policies/guardrails. The proposed framework shifts the paradigm to intentional sustainability. The framework provides implementable checkpoints that organizations can embed in their SDLC processes and offers researchers the opportunity to bridge the theory-practice gap, and highlights urgent needs for developer-centric sustainability tools and long-term effects on AI-induced technical debt.
Current research examines either technical benefits in isolated SDLC phases or the environmental footprint of AI systems in general. A holistic mapping of both positive (hand-prints) and negative (footprints) impacts across all phases and all five sustainability dimensions is missing, as is empirical evidence on real-world practitioner experiences and governance practices.
This thesis maps the sustainability impacts of AI in the SDLC and proposes a practical framework for its responsible integration. A mixed-methods approach was employed: 64 peer-reviewed studies (2020–2025) were used for a systematic literature review, and semi-structured interviews with 27 experienced software professionals from 12 countries representing diverse roles (junior to principal architects) and domains (financial services, telecommunications, aviation).
Key findings are: (1) AI use is heavily concentrated in the implementation and testing phases, with strong handprints in developer productivity, code quality, and testing efficiency. Still, it remains less in requirements engineering and maintenance despite academic recommendations. (2) Severe footprints are unmonitored energy consumption and carbon emissions, bias propagation, introduction of AI-specific bugs, and risk of developer skill degradation. (3) Practitioners prioritize immediate productivity gains and technical risks (security, code quality, technical debt) while remaining significantly less aware of environmental impacts than the academic literature suggests.
The results demonstrate that current AI adoption in software development achieves only “accidental” efficiency. This thesis outcome is a framework that consists of actionable components: mandatory training, human-in-the-loop oversight, and company policies/guardrails. The proposed framework shifts the paradigm to intentional sustainability. The framework provides implementable checkpoints that organizations can embed in their SDLC processes and offers researchers the opportunity to bridge the theory-practice gap, and highlights urgent needs for developer-centric sustainability tools and long-term effects on AI-induced technical debt.