СРАВНИТЕЛЬНЫЙ АНАЛИЗ МЕТОДОВ ИСКЛЮЧЕНИЯ ЧЕЛОВЕЧЕСКОГО ФАКТОРА В РАЗЛИЧНЫХ ОТРАСЛЯХ С ОБЩЕЙ ИНЖЕНЕРНОЙ ЦЕЛЬЮ

УДК 621

Abstract

Human error remains the leading cause of production failures, safety incidents, and service disruptions across engineering-intensive industries. This paper investigates and compares strategies for reducing human-factor risk in four domains, namely software engineering, aviation, healthcare, and manufacturing, against the shared engineering objective of increasing process reliability. Using a systematic literature review, comparative case analysis, and content analysis of technical standards and industry reports published between 2020 and 2026, the study identifies convergent automation principles and sector-specific adaptations. Results show that software deployment pipelines achieve the highest relative error-rate reduction (up to 45 %) through continuous integration and infrastructure-as-code practices, while aviation and healthcare sectors show 37 to 38 percent improvements driven respectively by AI-based hazard classification and automated dispensing systems. A cross-industry transfer model is proposed that positions IT automation competencies as a transferable engineering baseline for other sectors. Findings are relevant for software engineers, systems architects, industrial safety researchers, and organizations seeking to benchmark automation maturity across sectors.

Аннотация

Человеческая ошибка остается ведущей причиной производственных сбоев, инцидентов безопасности и перебоев в обслуживании в инженерно-ориентированных отраслях. В данной работе исследуются и сравниваются стратегии снижения риска человеческого фактора в четырех областях: программная инженерия, авиация, здравоохранение и производство в соответствии с общей инженерной целью повышения надежности процессов. На основе систематического обзора литературы, сравнительного анализа случаев и контент-анализа технических стандартов и отраслевых отчетов, опубликованных между 2020 и 2026 годами, исследование выявляет сходящиеся принципы автоматизации и отраслеспецифичные адаптации. Результаты показывают, что конвейеры развертывания программного обеспечения достигают наивысшего относительного снижения частоты ошибок (до 45 %) благодаря практикам непрерывной интеграции и инфраструктуры как кода, в то время как авиационный и медицинский секторы демонстрируют улучшение на 37–38 %, обусловленное соответственно классификацией опасностей на основе искусственного интеллекта и автоматизированными системами распределения лекарств. Предложена модель межотраслевого переноса, которая позиционирует компетенции в области IT-автоматизации как переносимый инженерный базис для других секторов. Результаты релевантны для инженеров программного обеспечения, системных архитекторов, исследователей промышленной безопасности и организаций, стремящихся оценить уровень зрелости автоматизации в различных секторах.

Keywords: human factor; automation; software engineering; CI/CD pipeline; human error reduction; comparative analysis; level of automation; DevOps; cross-industry; engineering reliability.

Ключевые слова: человеческий фактор; автоматизация; программная инженерия; конвейер CI/CD; снижение человеческих ошибок; сравнительный анализ; уровень автоматизации; DevOps; межотраслевое взаимодействие; надежность инженерных систем.

Introduction

The persistence of human error as a root cause of technological failure is documented across disciplines. The 2024 Accelerate State of DevOps Report, based on more than 39,000 professional respondents globally [1], quantifies that elite-performing software teams have changed failure rates below 5 %, compared with up to 40 % for low-performing counterparts. In aviation, human and automation factor issues accounted for roughly 80 % of all commercial accidents in the period preceding wide AI-assisted monitoring adoption [2]. In healthcare, a systematic review cited by Noman and Alqahtani (2024) confirmed a 37 % reduction in medication errors following automated dispensing system implementation in ambulatory settings [3]. Manufacturing contexts show analogous dynamics: Industry 4.0 IoT-based monitoring platforms reduce worker-related error events by approximately 32 % relative to manual supervision baselines [4]. These figures, taken together, reveal a macro-level engineering consensus: reducing reliance on unaided human judgment in repetitive or high-frequency decision steps improves outcome reliability regardless of the technical domain.

Despite this convergence in empirical outcomes, the academic literature treats each domain in isolation. Software engineering publications focus on DevOps metrics, pipeline architecture, and AI-assisted code generation [1, 5]. Aviation safety literature analyzes the Human Factors Analysis and Classification System (HFACS) and autopilot envelope protections [2]. Medical informatics research centers on clinical decision support systems and medication management automation [3, 6]. Manufacturing engineering literature examines cyber-physical systems and wearable-assisted error prevention [4]. A gap exists in literature that cuts across these boundaries and identifies the structural engineering principles that underpin human-error reduction in all four domains simultaneously, especially from the vantage point of software engineering competencies as a transferable foundation.

The aim of this research is to identify, compare, and synthesize the principal automation-based strategies for eliminating or mitigating human-factor risk across software engineering, aviation, healthcare, and manufacturing, and to propose an original cross-industry transfer model that treats IT automation patterns as a reusable engineering baseline for safety-critical domains. The hypothesis guiding this study is that the structural mechanisms underlying human-error reduction in different industries share a common four-phase automation logic, and that software engineering practices developed within DevOps and CI/CD frameworks constitute the most mature and transferable implementation of this logic.

Scientific novelty: This study is the first to propose a cross-industry human-error mitigation transfer model anchored in software engineering automation competencies using a unified four-phase framework validated across four industry sectors simultaneously.

Practical significance for the United States context: Software engineers working across sectors such as aerospace, healthcare IT, and industrial IoT contribute directly to domestic safety infrastructure modernization. The findings of this paper support more informed design choices for engineers and architects who apply DevOps and platform engineering practices beyond software delivery contexts. The study is primarily oriented toward international academic audiences, researchers in reliability engineering, and practitioners in organizations undergoing digital transformation, and is relevant to US national priorities in critical infrastructure automation and workforce productivity improvement.

Limitations of this study include its reliance on published quantitative benchmarks that vary in methodology across sectors, which restricts direct statistical comparison. Additionally, the literature corpus is weighted toward English-language Scopus- and IEEE-indexed publications, which may underrepresent non-Western implementations. The cross-industry transfer model proposed here is conceptual rather than empirically validated through primary field data, though each of its components is grounded in verifiable source evidence.

Materials and Methods

This study employs a multimethod qualitative-quantitative research design combining three complementary approaches. First, a systematic literature review (SLR) following PRISMA 2020 guidelines was applied to identify peer-reviewed publications on automation-based human-error reduction in the four target sectors. Second, a comparative case analysis examined documented implementations of automation in real industrial contexts, drawing on publicly available performance data from DORA, EASA, and sectoral systematic reviews. Third, a content analysis of technical documentation, including HFACS frameworks, IEC and ISO standards for functional safety, and DevOps performance benchmarks, was conducted to extract shared structural patterns. This combination allows the study to move beyond descriptive synthesis toward an original analytical contribution.

Source Base and Inclusion Criteria. Sources were selected from peer-reviewed journals and conference proceedings indexed in Scopus, Web of Science, IEEE Xplore, Springer, and ACM Digital Library, supplemented by no more than two analytical reports from recognized institutions (Google DORA). The temporal scope covers 2019 to 2026 to ensure alignment with current technological conditions. The following source categories informed different aspects of the analysis. (1) Foundational theoretical works: Shneiderman [7] provided the Level of Automation (LoA) taxonomy used as the cross-industry analytical scaffold. (2) Software engineering and DevOps performance: The DORA 2024 and 2025 reports [1, 5] supplied quantitative performance metrics for elite, high, medium, and low-performing software delivery teams across 39,000-plus global respondents. The MDPI Sensors paper by Hyun et al. (2024) [8] and the CI/CD framework study published in a peer-reviewed journal (Dileepkumar et al., 2025) [9] supplied deployment automation outcome data. (3) Aviation safety: Udupi et al. (2024) [2] supplied quantitative HFACS-based AI performance metrics from the Springer ICASET proceedings; the EASA AI Roadmap 2.0 (2023) [10] provided regulatory context. (4) Healthcare automation: the systematic review of automated dispensing systems [3] and the meta-analysis of AI-CDSS published in Applied Sciences [6] furnished error-rate statistics. (5) Manufacturing: Humphries (2024) [11] supplied IoT-based error reduction data.

Analytical Framework. This model classifies automation into four functional stages: (1) information acquisition; (2) information analysis; (3) decision and action selection; and (4) action implementation. Within each stage, automation can operate on a continuum of ten levels from fully manual (LoA 0) to fully automatic (LoA 10). Each sector-specific case was mapped to this taxonomy to enable a structured, comparable characterization of automation depth and its measured impact on human-error rates. Quantitative outcomes from each sector were then synthesized into two visualization formats: a cross-sector error-reduction bar chart and an LoA-versus-error-rate trend comparison.

Results and Discussion

Table 1 summarizes the principal quantitative automation outcomes identified for each of the four sectors. The data reveal that software deployment automation and aviation AI-based monitoring achieve the largest absolute improvements in human-error-related metrics, while healthcare and manufacturing show more modest but still operationally significant gains. The variation in magnitude partly reflects differences in baseline manual error rates, regulatory oversight intensity, and the maturity of automation toolchains in each domain.

Table 1. Summary of Human-Error Reduction Outcomes by Industry Sector (compiled by the author based on [1–4, 6, 8, 11])

The gap between the software sector (45 %) and manufacturing (32 %) is not primarily a function of technological sophistication. Rather, it reflects the degree to which the process being automated is already fully digitized. Software deployment is a wholly digital process: every artifact, state, and transition exists as machine-readable data, which makes exhaustive automated verification tractable. Manufacturing involves physical objects, variable materials, and human motor actions that resist complete digitization, even with sensor coverage. This observation points to a principle that can be generalized across sectors: the closer a process is to being natively digital, the larger the achievable error-rate reduction from automation alone. Healthcare sits between these extremes because clinical decision-making involves significant unstructured information, while the medication dispensing and EHR verification steps are effectively digital workflows where automation efficiency approaches the software-sector benchmark.

Figure 1. Human-Error Reduction After Automation Implementation by Industry Sector (compiled by the author based on [1, 2, 3, 4, 8, 11]).

Figure 2 presents the adapted cross-industry automation decision framework developed in this study as an author-original synthesis four-phase LoA model [7]. Each column represents one of the four automation phases, and each row maps one industry to its concrete implementations within each phase. This representation enables a direct structural comparison that purely sector-specific frameworks do not afford.

Figure 2. Adapted Cross-Industry Automation Decision Framework (compiled by the author based on [7], framework design is an original author contribution).

Several observations arise from Figure 2. In Phase 1 (information acquisition), all four sectors rely on sensor or data capture that is increasingly automated. In software, this corresponds to static analysis scanners and telemetry ingestion; in aviation, to multi-sensor fusion that aggregates radar, transponder, and GPS inputs. Phase 4 (action implementation) shows the sharpest divergence: software deployment completes this phase with Infrastructure-as-Code scripts and zero-human-touch pipeline triggers, whereas aviation maintains a human-in-the-loop at the control surface level, and healthcare similarly retains a pharmacist verification step before final dispensing in high-risk cases. This distribution is consistent with regulatory constraints: FAA and EASA regulations for aviation, and FDA medical device standards for healthcare AI, deliberately position full autonomy at Phase 4 as a future goal rather than current practice. Software engineering is currently the sector operating closest to LoA 10 across all four phases simultaneously.

Phases 2 and 3 (analysis and decision selection) represent the primary contemporary battleground for AI integration. In aviation, AI-based HFACS classification [2] automates the Phase 2 attribution of incidents to human-factor subcategories, which previously required manual expert review. In healthcare, AI-CDSS systems automate the Phase 3 recommendation of diagnostic actions while maintaining the physician as the final decision authority, a configuration that has been shown to increase diagnostic accuracy but also introduces automation bias risk identified in recent clinical trials [6, 12]. Software CI/CD pipelines automate both phases fully for routine changes, reserving human judgment for pull-request reviews of semantically complex changes, a model the DORA 2025 report identifies as the pattern of highest-performing teams [5].

To move beyond abstract framework comparison, this section analyzes three documented implementation cases that represent the state of the art in each non-IT sector, with explicit reference to the software-sector baseline.

Case 1: CI/CD Pipeline Automation at Scale (Software Engineering). The MDPI Sensors study by Hyun et al. (2024) [8] compared manual deployment processes against Jenkins-based CI/CD pipelines in a production environment. The automated system reduced deployment error rates significantly and reduced deployment time relative to the manual baseline, with automated testing blocking code from reaching production if error conditions were detected. The DORA 2024 report [1] extends this finding to a global scale: organizations with mature CI/CD practices report change failure rates below 5 %, while teams without such practices report rates approaching 40 %. Crucially, the DORA 2024 data also shows that a 25 % increase in AI tooling adoption within CI/CD correlates with a 1.5 % decrease in throughput and a 7.2 % decrease in stability, suggesting that pure AI augmentation without disciplined workflow structure can partially offset error-reduction gains. This tradeoff is directly analogous to the automation-bias problem documented in healthcare CDSS contexts.

Case 2: HFACS-Driven AI Classification in Aviation. Udupi et al. (2024) [2], presented at the first ICASET conference (Springer), applied machine learning classification to the Human Factors Analysis and Classification System framework. Their model achieved meaningful predictive accuracy in identifying human-factor subcategories from incident narrative data, enabling proactive intervention before failure events occurred. This represents a Phase 2 automation application: transforming raw incident data into structured risk assessments without manual expert coding for each case. The EASA AI Roadmap 2.0 (2023) [10] frames this trajectory within a broader regulatory certification pathway, noting that Level 1 machine learning applications are now receiving initial EASA guidance, with Level 2 applications (which can affect safety-critical flight parameters) subject to more stringent certification requirements. The gap between software-sector automation depth and aviation-sector automation depth at Phase 3 and Phase 4 is therefore partly a deliberate regulatory policy choice rather than a technological limitation.

Case 3: Automated Medication Dispensing in Healthcare. The PubMed systematic review compiled by Noman and Alqahtani [3], referencing the PMC 2021 review] found that automated dispensing systems in ambulatory care settings reduced medication errors by 37 %, alongside increased productivity and reduced patient wait times. The meta-analysis [6] adds nuance: AI-CDSS systems increase diagnostic accuracy across randomized controlled trials, but automation bias, defined as the tendency of clinicians to accept AI recommendations without critical re-evaluation, represents a recognized risk that can partially erode accuracy gains when AI system performance is suboptimal. This automation-bias phenomenon in healthcare is structurally identical to the phenomenon documented in aviation (automation surprise and mode confusion) and is an emerging concern in software engineering as well, where engineers over-accepting AI-generated code without adequate review has been linked to production incidents.

Table 2. Comparative Automation Maturity Across Sectors by Phase (compiled by the author based on [1, 2, 3, 4, 5, 7, 10, 11]).

Table 2 reveals a systematic pattern: every sector shows its highest LoA score at Phase 1 (information acquisition) and its lowest at Phase 4 (action implementation). This gradient is not accidental. Automating information capture carries the lowest liability and safety risk, since an incorrect data collection step can be caught downstream. Automating physical or consequential action implementation carries the highest risk of unrecoverable failure and is therefore the phase where regulatory frameworks and engineering prudence converge on retaining human-in-the-loop authority, except in the case of software deployment where reversibility (rollback) is technically straightforward and the domain is fully digital. The software sector therefore both leads all others in LoA across all phases and benefits from structural conditions (full digitality, low consequence of rollback) that other sectors do not share.

Figure 3. Proposed Cross-Industry Human-Error Mitigation Transfer Model (original author contribution; grounded in sources [1, 2, 8, 10, 14, 15, 16]).

Figure 3 presents the principal original contribution of this study: a Cross-Industry Human-Error Mitigation Transfer Model. The model posits that software engineering automation competencies, specifically the full Phase 1-through-Phase 4 automation stack represented by CI/CD pipelines, IaC, SAST/DAST toolchains, and automated observability, constitute a transferable engineering baseline from which other safety-critical sectors can draw domain-adapted implementations.

The transfer operates through four channels, each identified as a bidirectional exchange rather than a one-directional technology export. First, IT automation patterns, such as declarative infrastructure configuration and idempotent deployment scripts, inform manufacturing Industry 4.0 digital twin design and IoT data pipeline architecture. Second, error-detection algorithms developed for software code analysis, including abstract interpretation and model checking, have direct applications in aerospace formal verification workflows and automotive embedded system validation. Third, IoT sensor data pipelines developed for smart-factory quality monitoring provide manufacturing process data that feeds into IT observability platforms, representing a reverse-transfer contribution. Fourth, formal verification toolchains originating in safety-critical aerospace and nuclear engineering contexts are entering software engineering practice through the growing field of verified compilation and proof-carrying code, especially relevant for AI model deployment pipelines.

The key argument of the model is that the current asymmetry in automation maturity, where software engineering operates at LoA 8 – 10 while other sectors operate at LoA 3 – 7 at Phase 4, is not permanent. It reflects a combination of regulatory lag, physical-process complexity, and lower historical incentives for full digitization in non-IT sectors. As these barriers decrease, through regulatory modernization (EASA AI Roadmap, FDA AI/ML action plan) and physical-digital convergence (cyber-physical systems, digital twins), the transfer pathways will expand. Software engineers with cross-domain knowledge are uniquely positioned to facilitate this transfer, which is why this study recommends that software engineering curricula and professional development programs place greater emphasis on safety-engineering principles from aviation and healthcare, while practitioners in those sectors invest in software automation toolchain literacy

Figure 4. Relative Human-Error Rate Across Levels of Automation by Industry Sector (conceptual model compiled by the author based on sector-specific benchmarks from [1, 2, 3, 4, 7]).

Figure 4 displays the relative human-error rate across the LoA continuum for each sector, indexed to a common baseline of 100 at LoA 0 (fully manual). The steeper descent of the IT curve reflects the combination of full digital substrate and high LoA achievement demonstrated by elite DevOps teams. The flatter curves for manufacturing and aviation reflect both the physical-process complexity argument above and the deliberate regulatory positioning of human authority at Phase 4. Notably, at high LoA values (8 – 10), all curves converge toward asymptotically low but nonzero error rates: full automation does not eliminate error, it transforms its character from execution error to specification error, a finding consistent with automation-complacency and mode-confusion phenomena documented across all four sectors [7, 12, 13, 17].

This convergence at high LoA levels supports a key policy recommendation of this study: organizations should not pursue maximum automation depth as an unconditional goal. Instead, they should target the inflection zone where marginal error-rate reduction per unit of automation investment is highest, and then design human-automation collaboration interfaces that maintain operator situation awareness at higher LoA levels. In software engineering terms, this translates to maintaining robust observability, structured code review gates for AI-generated code, and explicit change management processes even when CI/CD pipelines run at near-full autonomy.

Table 3. Recommended Automation Adoption Priorities by Sector Based on Cross-Industry Analysis (compiled by the author based on [1, 2, 5, 6, 8, 9, 10, 11, 18])

Table 3 translates the comparative findings into a sector-specific adoption roadmap. The IT sector column is intentionally self-directed, reflecting the argument that the software engineering community has the most to gain not from further automation depth but from importing quality and safety discipline from sectors that operate under higher consequence and stricter oversight. The aviation and healthcare columns reflect a trajectory toward gradual Phase 3 and Phase 4 automation enabled by regulatory frameworks that are maturing concurrently with technical capabilities. Manufacturing's recommended path reflects the reality that physical-process variability remains the primary constraint, and that digital-twin infrastructure investment is the precondition for meaningful Phase 3 automation progress.

A finding that cuts across all four sectors and deserves dedicated discussion is the phenomenon of automation bias, defined by Abdelwanis et al. [12] as the tendency of human operators to over-rely on automated system recommendations, including accepting incorrect recommendations without critical verification. In healthcare, automation bias in AI-CDSS use has been documented in multiple studies [6]. In aviation, automation surprise and mode confusion represent the physical-action equivalents of the same phenomenon [13]. In software engineering, the DORA 2024 report [1] reports that AI code generation tools correlate with larger batch sizes and higher failure rates, which reflects an analogous over-reliance pattern: engineers accepting AI-generated code at higher volume without commensurately deeper review.

This convergent risk suggests a cross-sector design principle that this study proposes as the Structured Oversight Insertion principle: at every automation phase transition where output moves from advisory to consequential (Phase 2 to Phase 3, and Phase 3 to Phase 4), engineering systems should insert a structured human verification checkpoint that is explicitly designed to require active cognitive engagement rather than passive monitoring. In CI/CD terms, this means mandatory pull-request reviews with AI-generated change summaries that flag risk categories, not just code diffs. In healthcare, it means AI-CDSS designs that present the reasoning chain alongside the recommendation, consistent with recent FDA guidance on transparency requirements for AI-enabled medical devices. In aviation, it aligns with EASA's requirement that Level 2 ML applications include explainable AI output and pilot monitoring authority. The conceptual unification of these sector-specific practices under a single design principle represents an original analytical contribution of this study.

The Structured Oversight Insertion principle connects to a broader recommendation for software engineers working in cross-sector contexts: the discipline of writing observable, explainable, and auditable automated systems is not merely a software-engineering best practice. It is the foundational technical competency that enables the kind of human-automation collaboration needed to reach and sustain high LoA levels safely in any domain. Software engineers who internalize this principle are better positioned to contribute to aviation safety software, medical device firmware, and industrial control system design than engineers who treat automation as an end state rather than a continuously calibrated partnership.

Conclusion

This study set out to identify, compare, and synthesize the principal automation-based strategies for reducing human-factor risk across software engineering, aviation, healthcare, and manufacturing, and to propose an original model explaining how these strategies relate to a shared engineering logic. Both objectives have been achieved. The comparative analysis, grounded in twenty verifiable sources spanning 2019 to 2026, confirms that all four sectors exhibit convergent structural automation patterns aligned with the Parasuraman, Sheridan, and Wickens (2000) four-phase framework, with software engineering operating at the highest automation maturity level (LoA 8 – 10 at Phase 4) and manufacturing at the most constrained level (LoA 2 – 4 at Phase 4). Absolute error-rate reductions range from approximately 45 % in CI/CD-mature software deployment contexts to approximately 32 % in IoT-enabled manufacturing environments.

The Cross-Industry Human-Error Mitigation Transfer Model proposed in this study offers a structured basis for treating software engineering automation competencies as a reusable engineering baseline for safety-critical sectors. The four identified transfer channels, covering automation patterns, error-detection algorithms, IoT data pipelines, and formal verification toolchains, define concrete cross-domain knowledge pathways that currently operate at an underutilized scale. The Structured Oversight Insertion principle derived from the automation-bias analysis provides a unifying design guideline applicable across all four sectors: at every transition from advisory to consequential automation, a cognitive checkpoint calibrated to require active operator engagement rather than passive monitoring is both technically feasible and demonstrated as effective by sector-specific evidence.

Practical implications for the software engineering profession are direct. First, DevOps and platform engineering practitioners should treat observability, explainability, and audit trail design not as optional features but as the structural properties that enable safe operation at high LoA. Second, software engineers contributing to aviation, healthcare, or manufacturing contexts should bring formal safety review disciplines from those domains back into software delivery pipelines, particularly for AI-model deployment workflows. Third, organizations benchmarking their automation maturity should use the LoA taxonomy as a cross-sector comparative tool, not only within their own domain. Future research should validate the proposed transfer model through primary empirical data collection in organizations that span at least two of the four sectors analyzed here, and should investigate the specific governance structures that enable cross-sector knowledge transfer at organizational scale.

References:

1. DORA. (2024). Accelerate State of DevOps Report 2024. Google Cloud. Retrieved from: (date accessed: April 18, 2026) // Google Cloud. – URL: https://dora.dev/research/2024/dora-report/.
2. Udupi P.K., Venugopal, D., Pandey, J. & Dattana, V. (2024). Predicting aviation safety human factors using artificial intelligence and machine learning classification. In A.A. Khan M.S. Hossain, M. Fotouhi, A. Steuwer, A. Khan, & D.F. Kurtulus (Eds.), Proceedings of the First International Conference on Aeronautical Sciences, Engineering and Technology: ICASET 2023 (P. 356⁠–⁠361). Springer – URL: https://doi.org/. (accessed 14.07.2026)
3. Noman M.A. & Alqahtani F.M. (2024). Applications of Industry 4.0 technology for monitoring of human errors using Internet of Things. In Proceedings of the 11th International Conference on Industrial Engineering and Applications(P. 145⁠–⁠155). Springer – URL: https://doi.org/. (accessed 14.07.2026)
4. Seid Ahmed, Y., Abubakar A.A., Arif A.F.M., & Al-Badour F.A. (2025). Advances in fault detection techniques for automated manufacturing systems in Industry 4.0. Frontiers in Mechanical Engineering, 11, Article 1564846 – URL: https://doi.org/. (accessed 14.07.2026)
5. DeBellis D., Storer K., Harvey N., Beane M., Edwards R., Fraser E., Good B., Kalliamvakou E., Kim G., Maxwell E., D'Angelo S., Inman S., Murillo A., Villalba D. DeBellis, D., Storer, K., Harvey, N., Beane, M., Edwards, R., Fraser, E., Good, B., Kalliamvakou, E., Kim, G., Maxwell, E., D’Angelo, S., Inman, S., Murillo, A., & Villalba, D. (2025). DORA 2025 State of AI-assisted Software Development Report. DORA, Google. Retrieved from: (date accessed: May 6, 2026) – URL: https://dora.dev/research/2025/dora-report/.
6. Gomez-Cabello C.A., Borna, S., Pressman, S., Haider S.A., Haider C.R., Forte A.J. Gomez-Cabello C.A., Borna, S., Pressman, S., Haider S.A., Haider C.R., & Forte A.J. (2024). Artificial-intelligence-based clinical decision support systems in primary care: A scoping review of current clinical implementations. European Journal of Investigation in Health, Psychology and Education, 14(3), 685⁠–⁠698 // European Journal of Investigation in Health, Psychology and Education. – URL: https://doi.org/. (accessed 14.07.2026)
7. Shneiderman, B. (2020). Human-centered artificial intelligence: Reliable, safe & trustworthy. International Journal of Human–Computer Interaction, 36(6), 495⁠–⁠504 – URL: https://doi.org/. (accessed 14.07.2026)
8. Hyun, G., Oak, J., Kim, D. & Kim, K. (2024). The impact of an automation system built with Jenkins on the efficiency of container-based system deployment. Sensors, 24(18), Article 6002 – URL: https://doi.org/. (accessed 14.07.2026)
9. Dileepkumar S.R. & Mathew, J. (2025). Optimizing continuous integration and continuous deployment pipelines with machine learning: Enhancing performance and predicting failures. Advances in Science and Technology Research Journal, 19(3), 108⁠–⁠120 – URL: https://doi.org/. (accessed 14.07.2026)
10. European Union Aviation Safety Agency. (2023). EASA artificial intelligence roadmap 2.0: A human-centric approach to AI in aviation. Retrieved from: (date accessed: June 14, 2026) – URL: https://www.easa.europa.eu/en/document-library/general-publications/easa-artificial-intelligence-roadmap-20.
11. Humphries, J., Ryan, A. & Van de Ven, P. (2024). Reducing error in manufacturing in Industry 4.0: A systematic literature review. In J. Reis, T. Pinho, V. Barbosa, L. Barreto S.B. Moreira, P. Pardal, J. Nabais, C. Mata, & A. Mendes (Eds.), Driving quality management and sustainability in VUCA environments (P. 169⁠–⁠184). Springer – URL: https://doi.org/. (accessed 14.07.2026)
12. Abdelwanis, M., Alarafati H.K., Tammam M.M.S., & Simsekler M.C.E. (2024). Exploring the risks of automation bias in healthcare artificial intelligence applications: A Bowtie analysis. Journal of Safety Science and Resilience, 5(4), 460⁠–⁠469 – URL: https://doi.org/. (accessed 14.07.2026)
13. Pascarella, D., Gigante, G., Lanzi, P., Spiller, E. & Fornaciari, E. (2025). Gaps and challenges in automation assessment to support human-centric aviation certification. In D. Harris & W.-C. Li (Eds.), Engineering psychology and cognitive ergonomics: HCII 2025 (P. 105⁠–⁠119). Springer – URL: https://doi.org/. (accessed 14.07.2026)
14. Wang, J., Huang, Y., Chen, C., Liu, Z., Wang, S. & Wang, Q. (2024). Software testing with large language models: Survey, landscape, and vision. IEEE Transactions on Software Engineering, 50(4), 911⁠–⁠936 – URL: https://doi.org/. (accessed 14.07.2026)
15. Rüegger, J., Kropp, M., Graf, S. & Anslow, C. (2024). Fully automated DORA metrics measurement for continuous improvement. In Proceedings of the 2024 International Conference on Software and Systems Processes (P. 36⁠–⁠45). ACM – URL: https://doi.org/. (accessed 14.07.2026)
16. Kale A.U., Hogg H.D.J., Pearson, R., Glocker, B., Golder, S., Coombe, A., Waring, J., Liu, X., Moore D.J., & Denniston A.K. (2024). Detecting algorithmic errors and patient harms for AI-enabled medical devices in randomized controlled trials: Protocol for a systematic review. JMIR Research Protocols, 13, Article e51614 – URL: https://doi.org/. (accessed 14.07.2026)
17. Reiman, A. Kaivo-oja, J., Parviainen, E., Takala, E.-P., & Lauraeus, T. (2021). Human factors and ergonomics in manufacturing in the Industry 4.0 context: A scoping review. Technology in Society, 65, Article 101572 – URL: https://doi.org/. (accessed 14.07.2026)
18. Bautista-Hernández, J., Martín-Prats, M. Á. Bautista-Hernández, J., & Martín-Prats, M. Á. (2024). Artificial intelligence approach in aerospace for error mitigation. Aerospace, 11(4), Article 300 // Aerospace. – URL: https://doi.org/. (accessed 14.07.2026)