pith. sign in

arxiv: 2604.09633 · v1 · submitted 2026-03-19 · 💻 cs.CY · cs.AI

Agentic AI in Engineering and Manufacturing: Industry Perspectives on Utility, Adoption, Challenges, and Opportunities

Kristen M. Edwards , Maxwell Bauer , Claire Jacquillat , A. John Hart , Faez Ahmed This is my paper

Pith reviewed 2026-05-15 07:51 UTC · model grok-4.3

classification 💻 cs.CY cs.AI
keywords agentic AIengineering workflowsmanufacturing adoptiondata fragmentationlegacy toolchainshuman-in-the-loopregulatory barriersqualitative interviews
0
0 comments X

The pith

Agentic AI adoption in engineering and manufacturing is limited more by fragmented data, security rules, and legacy toolchains than by model capabilities.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper reports on an interview-based study of how agentic AI is entering engineering and manufacturing work. Gains today appear mostly in repetitive, structured tasks, with larger potential in agents that coordinate steps across different software tools. The main obstacles are not weak AI models but messy data that machines cannot easily read, strict security and regulatory demands, and old design tools that lack open connections. Progress is expected to move in stages, starting with low-stakes help and advancing to more automation only after verification methods and trust improve. The work also notes organizational gaps such as uneven AI knowledge inside companies and governance that has not adapted to agentic systems.

Core claim

Near-term AI gains cluster around structured, repetitive work and data-intensive synthesis, while higher-value agentic gains come from orchestrating multi-step workflows across tools. Adoption is constrained less by model capability than by fragmented and machine-unfriendly data, stringent security and regulatory requirements, and limited API-accessible legacy toolchains. Reliability, verification, and auditability are central requirements for adoption, driving human-in-the-loop frameworks and governance aligned with existing engineering reviews. Beyond technical barriers there are also organizational ones: a persistent AI literacy gap, cultural heterogeneity, and governance structures that

What carries the argument

The staged progression of AI utility from low-consequence assistance toward higher-order automation, as trust, infrastructure, and verification mature.

If this is right

  • AI deployment will succeed first in low-risk, repetitive tasks before expanding to multi-tool workflow orchestration.
  • Human oversight and verification processes will remain necessary for any higher-stakes use.
  • Technical progress requires integration with existing engineering data formats and legacy software APIs.
  • Companies will need parallel investments in AI literacy and updated governance to match agentic capabilities.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Standardizing engineering data formats and exposing more legacy tools through APIs could unlock faster adoption than further model scaling.
  • The same infrastructure and regulatory barriers may appear in other regulated sectors such as aerospace or medical device design.
  • A quantitative follow-up study measuring the relative weight of data versus model factors across firm sizes would test the interview findings at scale.

Load-bearing premise

The sample of over 30 interviews across four stakeholder groups is sufficiently representative to support general statements about industry-wide barriers and opportunities.

What would settle it

A broader survey of hundreds of engineering and manufacturing firms that ranks AI model limitations or insufficient reasoning ability as the primary barrier to adoption would undermine the central claim.

Figures

Figures reproduced from arXiv: 2604.09633 by A. John Hart, Claire Jacquillat, Faez Ahmed, Kristen M. Edwards, Maxwell Bauer.

Figure 1
Figure 1. Figure 1: Overall findings for agentic AI adoption in engineering and manufacturing. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: U.S. manufacturing employment in thousands of people. Source of data: U.S. Bureau of Labor [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: large engineering or manufacturing enterprises, their small or medium counterparts, AI developers, [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Examples of the various sources of data in engineering and manufacturing workflows. [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
read the original abstract

This work examines how AI, especially agentic systems, is being adopted in engineering and manufacturing workflows, what value it provides today, and what is needed for broader deployment. This is an exploratory and qualitative state-of-practice study grounded in over 30 interviews across four stakeholder groups (large enterprises, small/medium firms, AI developers, and CAD/CAM/CAE vendors). We find that near-term AI gains cluster around structured, repetitive work and data-intensive synthesis, while higher-value agentic gains come from orchestrating multi-step workflows across tools. Adoption is constrained less by model capability than by fragmented and machine-unfriendly data, stringent security and regulatory requirements, and limited API-accessible legacy toolchains. Reliability, verification, and auditability are central requirements for adoption, driving human-in-the-loop frameworks and governance aligned with existing engineering reviews. Beyond technical barriers there are also organizational ones: a persistent AI literacy gap, cultural heterogeneity, and governance structures that have not yet caught up with agentic capabilities. Together, the findings point to a staged progression of AI utility from low-consequence assistance toward higher-order automation, as trust, infrastructure, and verification mature. This highlights key breakthroughs needed, including integration with traditional engineering tools and data types, robust verification frameworks, and improved spatial and physical reasoning.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. This paper conducts an exploratory qualitative study based on over 30 interviews with stakeholders from large enterprises, small/medium firms, AI developers, and CAD/CAM/CAE vendors to examine the adoption of agentic AI in engineering and manufacturing. It claims that current utilities are in structured tasks, adoption barriers are primarily data, security, and legacy systems rather than model capabilities, and highlights needs for verification, human oversight, and organizational adaptations.

Significance. The study provides grounded industry perspectives on an emerging topic, with strengths in its multi-stakeholder sampling and focus on practical constraints. If the descriptive findings are accurate, they offer useful guidance for prioritizing integration efforts and governance in AI for manufacturing.

major comments (2)
  1. The sampling strategy and criteria for selecting the over 30 interviewees across the four groups are not sufficiently detailed (Methods section), which is important for evaluating the representativeness of the reported perspectives on industry-wide barriers.
  2. Some synthesized claims, such as the staged progression of AI utility, would benefit from more explicit linkage to specific interview excerpts or thematic codes (Results section) to strengthen the evidential basis.
minor comments (2)
  1. Consider adding the exact number of interviews and a brief breakdown by stakeholder group to enhance precision (Abstract).
  2. The term 'agentic AI' is used throughout but could include a concise definition in the introduction for broader accessibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their positive assessment and constructive feedback on our manuscript. We address each major comment below.

read point-by-point responses

  1. Referee: The sampling strategy and criteria for selecting the over 30 interviewees across the four groups are not sufficiently detailed (Methods section), which is important for evaluating the representativeness of the reported perspectives on industry-wide barriers.

    Authors: We agree that the sampling strategy requires more detail to allow evaluation of representativeness. In the revised manuscript, we have expanded the Methods section to specify the selection criteria (including minimum professional experience, role diversity, and company size categories), the recruitment approach through industry networks and events, and how we ensured coverage across the four stakeholder groups. revision: yes

  2. Referee: Some synthesized claims, such as the staged progression of AI utility, would benefit from more explicit linkage to specific interview excerpts or thematic codes (Results section) to strengthen the evidential basis.

    Authors: We appreciate this recommendation for enhancing the evidential basis. We have revised the Results section to include explicit linkages, such as referencing specific thematic codes (e.g., 'utility_structured_tasks') and incorporating additional interview excerpts that directly support the description of the staged progression of AI utility from low-consequence assistance to higher-order automation. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper presents an exploratory qualitative study grounded in over 30 interviews across stakeholder groups. All central claims are framed as direct descriptive findings from those responses, with no mathematical derivations, parameter fitting, equations, or self-referential reductions. No load-bearing self-citations, ansatzes, or renamings of known results appear; the methodology is self-contained and follows standard qualitative practices for state-of-practice reports.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claims rest on the assumption that stakeholder interviews accurately capture industry realities without independent corroboration from quantitative metrics or broader surveys.

axioms (1)
  • domain assumption Stakeholder interview responses provide a valid and representative view of current AI adoption practices and barriers.
    The study is entirely grounded in self-reported qualitative data from the 30+ interviews.

pith-pipeline@v0.9.0 · 5546 in / 1199 out tokens · 51373 ms · 2026-05-15T07:51:46.288683+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

4 extracted references · 4 canonical work pages

  1. [1]

    Ben Armstrong, Valerie K

    Accessed: 2026-03-02. Ben Armstrong, Valerie K. Chen, Alex Cuellar, Alexandra Forsey-Smerek, and Julie Shah. Automation from the worker’s perspective: How can new technologies make jobs better? Technical report, MIT Work of the Future / MIT Industrial Performance Center, 2024. CAD-Coder: An Open-Source Vision-Language Model for Computer-Aided Design Code ...

    work page doi:10.1115/detc2025-169758 2026

  2. [2]

    IEEE Future Directions

    Accessed: 2025-11-18. IEEE Future Directions. Ieee technology megatrends 2024, 2024. URL https://futuredirections.ieee.org/tech-megatrends/. IEEE Future Directions Report, accessed 2025-11-11. Heewoo Jun and Alex Nichol. Shap-e: Generating conditional 3d implicit functions, 2023. Thomas A. Kochan, Ben Armstrong, Julie Shah, Emilio J. Castilla, Ben Likis, ...

    work page 2025

  3. [3]

    URLhttps://arxiv.org/abs/2506.16504. Iain J. Marshall and Byron C. Wallace. Toward systematic review automation: A practical guide to using machine learning tools in research synthesis.Systematic Reviews, 8(1):163, 2019. ISSN 2046-4053. doi: 10.1186/s13643-019-1074-9. URLhttps://doi.org/10.1186/s13643-019-1074-9. Maxim Massenkoff and Peter McCrory. Labor ...

    work page doi:10.1186/s13643-019-1074-9 2019

  4. [4]

    Artificial Intelligence Review , author =

    Accessed: 2025-11-18. National Association of Manufacturers. Facts about manufacturing: The top 18 facts you need to know. https://nam.org/mfgdata/facts-about-manufacturing-expanded/, 2025. Accessed: 2025-11-11. Benjamin Nuernberger, Anny Liu, Heather Stefanini, Richard Otis, Amanda Towler, and R. Peter Dil- lon. AiBAT: Artificial Intelligence/Instruction...

    work page doi:10.1007/s10462-025-11290-y 2025