EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

From Automation to Autonomy: A Survey of Agentic Workflows in CI/CD Orchestration

Ali Amar, Ibrahim Qaiser, Ayesha Kanwal

International Journal of Innovations in Science and Technology · 14.05.2026 · 476

DOI: 10.33411/ijist/1813 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Innovations in Science and Technology, online erschienen am 14.05.2026. Verfasst von Ali Amar und 2 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Continuous Integration and Continuous Deployment (CI/CD) pipelines are foundational to modern software delivery, yet remain reliant on rigid, pre-defined scripts that lack the flexibility to handle unforeseen anomalies. While prior surveys have examined Large Language Model (LLM) agents for general software engineering tasks, none have focused specifically on CI/CD orchestration and the transition from automation to autonomy. Through a structured review of 72 papers published between 2023 and 2025 and sourced from IEEE Xplore, ACM Digital Library, and arXiv, this survey addresses that gap. The 72 studies are distributed across three primary application domains: autonomous code generation and repair (28 papers), intelligent verification and environment setup (23 papers), and incident management with root cause analysis (21 papers). We propose the PARA (Perception, Action, Reasoning, Reflection) framework as an operational lens for analyzing agentic CI/CD systems. Comparative analysis of five representative systems yields concrete performance figures: SWE-agent resolves 12.5% of issues against a 3.8% scripted baseline; the DEI multi-agent committee reaches 34.3% versus 27.3% for single-agent baselines; CXXCrafter achieves 71.2% success on C/C++ builds compared with 45% for general-purpose agents; MACOG reaches 74.02% on Terraform synthesis, dropping to 61.45% when its Security Prover is ablated; and Flow reports a 67% reduction in Mean Time to Resolution for incident triage. Reported task success across the 72 studies ranges from 15% to 75% as a function of task complexity, and self-correction loops add a further 4–5% per iteration at exponential token cost. Challenges spanning economic viability, security risks, and reliability concerns are systematically analyzed. We conclude that the shift from scripted automation to autonomous agents represents a significant evolution in DevOps practices toward intent-driven orchestration, and outline future directions, including Knowledge Graph-augmented LLMs and standardized Agent-Tool Protocols.

Zitieren

APA

Amar, A., Qaiser, I., & Kanwal, A. (2026). From Automation to Autonomy: A Survey of Agentic Workflows in CI/CD Orchestration. International Journal of Innovations in Science and Technology, 476. https://doi.org/10.33411/ijist/1813

Vancouver

Amar A, Qaiser I, Kanwal A. From Automation to Autonomy: A Survey of Agentic Workflows in CI/CD Orchestration. International Journal of Innovations in Science and Technology. 2026:476. doi:10.33411/ijist/1813

BibTeX
@article{Amar2026,
  author = {Amar, Ali and Qaiser, Ibrahim and Kanwal, Ayesha},
  title = {From Automation to Autonomy: A Survey of Agentic Workflows in CI/CD Orchestration},
  journal = {International Journal of Innovations in Science and Technology},
  year = {2026},
  pages = {476},
  doi = {10.33411/ijist/1813}
}
Kostenlos starten →