<![CDATA[The adoption of Artificial Intelligence (AI) systems demands an efficient and flexible development approach to address evolving data and requirements. This study conducted a PRISMA 2020–based Systematic Literature Review (SLR) comparing Agile and Waterfall approaches in AI development, focusing on (RQ1) time/throughput-based and quality-related efficiency and (RQ2) adaptability to change. Searches covered publications from 2021 to 2026 across five databases, followed by stepwise screening, full-text assessment, and methodological quality appraisal. Eight studies were retained: five explicitly addressing AI/MLOps contexts and three related software-engineering studies used as contextual, transferable evidence. The studies were analyzed through narrative synthesis using predefined operational indicators for efficiency and adaptability, with greater interpretive weight given to higher-quality, AI-specific studies. The evidence indicates that Agile, particularly when aligned with MLOps practices (e.g., CI/CD/CT, pipeline automation, monitoring, and retraining), is often associated with faster iterations and better responsiveness to data changes. Waterfall remains relevant in settings with strong governance, relatively stable requirements, and strict documentation, traceability, and auditability needs. A hybrid approach can balance structured control with iterative adaptation in complex or multi-domain AI projects. Given the small number of included studies (n=8) and the partial reliance on related software engineering evidence, these findings are indicative and context-dependent rather than conclusive; the choice of approach is shaped by the project's level of uncertainty, data dynamics, and the organization's AI/MLOps maturity.]]>
Copyrights © 2026
