The rapid deployment of AI agents and agentic AI systems in manufacturing has produced a growing body of research, yet it remains unclear whether implementation patterns, reported benefits, and dominant barriers identified in one manufacturing context apply to another. This thesis addressed that gap through a structured literature review of 90 peer-reviewed studies published between 2019 and 2026, examining AI agent and agentic AI implementations across eight manufacturing sectors: automotive, electronics and semiconductor, aerospace, chemical and process, energy, food and beverage, textile and apparel, pharmaceutical, and one technology-defined additive-manufacturing domain. Architectural categories were identified across the corpus, including rule-based and reinforcement-learning agents on the AI agent side, and LLM-enabled, knowledge-graph-integrated, digital-twin-integrated, and vision-language systems on the agentic AI side. Hybrid architectures, which combine multiple agent types, accounted for 34.4% of the corpus. Adoption maturity was highest in automotive and semiconductor manufacturing and lowest in textile, food and beverage, and additive manufacturing, reflecting differences in digital infrastructure, regulatory environment, and research investment rather than absence of applicable technology. Outcome measurement was broadly positive in direction but inconsistent across studies, with productivity, quality, and efficiency defined and measured differently. Three barrier categories were identified across all sectors: technical, organisational and regulatory, and scaling. The principal implication is that barrier diagnosis should precede deployment strategy, because the constraint most likely to prevent scale-up differs by sector and an intervention calibrated to the wrong barrier will not advance deployment. The thesis contributes a cross-sector comparative framework and an integrated sector-differentiated barrier-analysis framework as diagnostic instruments for researchers and practitioners seeking to close the gap between demonstrated technical capability and sustained industrial deployment.
The rapid deployment of AI agents and agentic AI systems in manufacturing has produced a growing body of research, yet it remains unclear whether implementation patterns, reported benefits, and dominant barriers identified in one manufacturing context apply to another. This thesis addressed that gap through a structured literature review of 90 peer-reviewed studies published between 2019 and 2026, examining AI agent and agentic AI implementations across eight manufacturing sectors: automotive, electronics and semiconductor, aerospace, chemical and process, energy, food and beverage, textile and apparel, pharmaceutical, and one technology-defined additive-manufacturing domain. Architectural categories were identified across the corpus, including rule-based and reinforcement-learning agents on the AI agent side, and LLM-enabled, knowledge-graph-integrated, digital-twin-integrated, and vision-language systems on the agentic AI side. Hybrid architectures, which combine multiple agent types, accounted for 34.4% of the corpus. Adoption maturity was highest in automotive and semiconductor manufacturing and lowest in textile, food and beverage, and additive manufacturing, reflecting differences in digital infrastructure, regulatory environment, and research investment rather than absence of applicable technology. Outcome measurement was broadly positive in direction but inconsistent across studies, with productivity, quality, and efficiency defined and measured differently. Three barrier categories were identified across all sectors: technical, organisational and regulatory, and scaling. The principal implication is that barrier diagnosis should precede deployment strategy, because the constraint most likely to prevent scale-up differs by sector and an intervention calibrated to the wrong barrier will not advance deployment. The thesis contributes a cross-sector comparative framework and an integrated sector-differentiated barrier-analysis framework as diagnostic instruments for researchers and practitioners seeking to close the gap between demonstrated technical capability and sustained industrial deployment.
AI Agents and Agentic AI in Manufacturing: A Cross-Industry Comparative Analysis of Implementation, Challenges, and Outcomes
TUNGALAG, DOLGION
2025/2026
Abstract
The rapid deployment of AI agents and agentic AI systems in manufacturing has produced a growing body of research, yet it remains unclear whether implementation patterns, reported benefits, and dominant barriers identified in one manufacturing context apply to another. This thesis addressed that gap through a structured literature review of 90 peer-reviewed studies published between 2019 and 2026, examining AI agent and agentic AI implementations across eight manufacturing sectors: automotive, electronics and semiconductor, aerospace, chemical and process, energy, food and beverage, textile and apparel, pharmaceutical, and one technology-defined additive-manufacturing domain. Architectural categories were identified across the corpus, including rule-based and reinforcement-learning agents on the AI agent side, and LLM-enabled, knowledge-graph-integrated, digital-twin-integrated, and vision-language systems on the agentic AI side. Hybrid architectures, which combine multiple agent types, accounted for 34.4% of the corpus. Adoption maturity was highest in automotive and semiconductor manufacturing and lowest in textile, food and beverage, and additive manufacturing, reflecting differences in digital infrastructure, regulatory environment, and research investment rather than absence of applicable technology. Outcome measurement was broadly positive in direction but inconsistent across studies, with productivity, quality, and efficiency defined and measured differently. Three barrier categories were identified across all sectors: technical, organisational and regulatory, and scaling. The principal implication is that barrier diagnosis should precede deployment strategy, because the constraint most likely to prevent scale-up differs by sector and an intervention calibrated to the wrong barrier will not advance deployment. The thesis contributes a cross-sector comparative framework and an integrated sector-differentiated barrier-analysis framework as diagnostic instruments for researchers and practitioners seeking to close the gap between demonstrated technical capability and sustained industrial deployment.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tungalag_Dolgion.pdf
accesso aperto
Dimensione
2.32 MB
Formato
Adobe PDF
|
2.32 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/118966