This thesis, submitted in April 2025 for the degree of Especialista en Ingeniería de Software (English title rendered here as “From Obstacles to Strategies: Integrating Artificial Intelligence into the Software Development Life Cycle”), asks what the main challenges and barriers are in integrating AI into the software development life cycle (SDLC). It is a narrative review: articles published between 2022 and 2025, found mainly through Google Scholar with IEEE Xplore, Scopus, ACM Digital Library and Web of Science also consulted, were filtered by inclusion and exclusion criteria, scored in a quality matrix, and coded into a data-extraction matrix of 30 articles recording, per SDLC phase, the challenge, its contributing factor, the suggested strategy and the reported impact.
The review scopes itself to integrating AI into traditional software-development environments, and much of the literature it synthesizes concerns machine learning, NLP and predictive models alongside generative assistants such as GitHub Copilot and ChatGPT.
Key Points
- AI tools are classified by SDLC phase — planning, requirements, design, implementation, testing, deployment and maintenance — from effort-estimation models and NLP requirements analysis to code generators, AI test frameworks, AI-driven CI/CD and predictive monitoring.
- The challenges found are grouped into three dimensions: technical (dependence on clean, structured data; “black box” opacity; incompatibility with legacy systems; errors and vulnerabilities from over-reliance on code generators), organizational (resistance to change, fear of losing roles or skills, lack of training, limited management support) and ethical (data privacy and security, transparency and explainability).
- From the matrix, the review identifies planning, testing and quality assurance, and deployment and maintenance as the SDLC phases where integrating AI is most critical.
- It reports that the literature addresses technical challenges in detail but is fragmented and less homogeneous in proposing solutions that combine the technical, organizational and ethical dimensions, and calls for a holistic framework.
- For code generation it concludes that automation needs to be combined with human review to ensure the correctness and robustness of the code.
- Its recommendations are grouped as technical (data curation, explainable AI, interoperable platforms, modular architectures), organizational (continuous training, interdisciplinary teams, management commitment, an open learning culture) and ethical (privacy policies, audits, transparency in automated decisions).
- In its quality assessment it reports that most reviewed articles define clear objectives but often lack the technical detail needed for reproducibility, and that most concentrate on DevOps, testing and development rather than planning and maintenance.
Notes
The author acknowledges the limits of a narrative review: the selection of articles reflects the author’s own decisions, the search did not follow a fully systematic protocol, and relevant studies may have been omitted, which limits generalization. The thesis recommends complementing it with systematic reviews and further empirical validation of the proposed strategies.