EUVIMEDEUVIMED entdecken
Systematische Übersicht

Artificial intelligence in the software development life cycle: A systematic review of operational efficiency evidence (2023–2026)

Ivan Okuskov

Multidisciplinary Reviews · 28.10.2026 · 10 (4) · 2027230

DOI: 10.31893/multirev.2027230 ↗

Kurz gefasst

Systematische Übersicht in der Zeitschrift Multidisciplinary Reviews, online erschienen am 28.10.2026. Verfasst von Ivan Okuskov. Die Arbeit nennt 53 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The adoption of artificial intelligence is radically transforming software development: large language models, deep learning architectures, and ML classifiers are being applied across all phases of the software development life cycle (SDLC). However, evidence of AI operational efficiency remains fragmented, which hinders practitioners from making informed adoption decisions. The present systematic review consolidates the evidence base in accordance with the PRISMA 2020 protocol. The search was conducted in May 2026 in Scopus and Google Scholar using a structured query restricted to the title field and covering the period 2023–2026. Following a two-stage screening process, 52 peer-reviewed studies from 22 countries were included in the final corpus out of 58 assessed for full-text eligibility. The corpus spans a spectrum of techniques — from classical ML classifiers and deep learning architectures to generative LLMs (GPT, Claude, Gemini, DeepSeek). The synthesis is organized across ten analytical SDLC clusters — from requirements engineering to software maintenance, with separate treatment of DevOps and security. The review addresses four research questions (RQ1–RQ4): AI techniques employed, operational efficiency gains achieved, metric comparability, and priority gaps. The most mature evidence base has emerged in code generation: structured prompting and verification-based architectures yield correctness gains of 6–27% with a 50.6% reduction in the development cycle. In testing, ensemble ML methods consistently demonstrate F1 scores above 0.97 in defect prediction. The multiplicative effect of requirements quality is significant: the precision of requirements formalization directly determines the effectiveness of subsequent implementation and verification phases. Systematic gaps have been identified across four directions: the absence of standardized delivery metrics (DORA), a deficit of longitudinal field studies, insufficient cross-domain validation, and limited understanding of human–AI interaction. The findings form an evidence map of priority AI adoption points and a research agenda for next-generation systems.

Zitieren

APA

Okuskov, I. (2026). Artificial intelligence in the software development life cycle: A systematic review of operational efficiency evidence (2023–2026). Multidisciplinary Reviews, 10(4), 2027230. https://doi.org/10.31893/multirev.2027230

Vancouver

Okuskov I. Artificial intelligence in the software development life cycle: A systematic review of operational efficiency evidence (2023–2026). Multidisciplinary Reviews. 2026;10(4):2027230. doi:10.31893/multirev.2027230

BibTeX
@article{Okuskov2026,
  author = {Okuskov, Ivan},
  title = {Artificial intelligence in the software development life cycle: A systematic review of operational efficiency evidence (2023–2026)},
  journal = {Multidisciplinary Reviews},
  year = {2026},
  volume = {10},
  number = {4},
  pages = {2027230},
  doi = {10.31893/multirev.2027230}
}
Kostenlos starten →