Pith. sign in

REVIEW 4 major objections 6 minor 103 references

This paper proposes a task-level Autonomy Score for agentic AI, and argues that in a simulated supply chain higher autonomy helps upstream tiers while raising costs for downstream tiers.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 02:30 UTC pith:DP36YA7V

load-bearing objection Plausible task-level autonomy metric, but the headline empirical claim rests on unreported weights and a very thin experiment. the 4 major comments →

arxiv 2607.25405 v1 pith:DP36YA7V submitted 2026-07-28 cs.HC

Agentic AI Autonomy Assessment: A Decision-Support Framework Towards Governed Supply Chain Systems

Lennart Trumpler , Rodrigo Furlan de Assis , Elias Ribeiro da Silva , Luis Antonio de Santa-Eulalia , Christian Hendriksen This is my paper
classification cs.HC
keywords agentic AIautonomy measurementsupply chain governancehuman-AI collaborationmulti-agent systemsbeer distribution gamelarge language modelsdecision support
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper proposes the Agentic AI Autonomy Assessment (AAAA) framework, which measures how autonomous an agentic AI system is at any moment by counting task-level interactions rather than classifying systems into discrete levels. It defines autonomy through two observable quantities—how many executable tasks the agent generates itself (initiative) and how often it pauses to consult an external decision-maker (decision control)—and combines them into a single Autonomy Score. The authors argue this score is objective, implementation-independent, and monitorable from development through runtime to end-of-life, making it a governance instrument rather than a performance metric. To test the construct, they run an agentic beer distribution game across five prompt-defined autonomy levels and find that higher autonomy lowers costs for upstream factory and distributor tiers but raises costs for the retailer, with a pooled slope coefficient showing the cost gradient increases with each tier. The upshot is that autonomy should be a deliberately calibrated design and governance variable, not something to be maximized.

Core claim

The central claim is that autonomy in agentic systems is not a capability or a discrete class but an observable property of how tasks are distributed between the agent and external decision-makers. The framework quantifies this through the Autonomy Score, AS = w1·IR + w2·(1−CR), where IR is the share of tasks generated internally through planning and goal decomposition and CR is the share of tasks involving consultation before execution. Weights allow context-specific balancing of initiative against decision control. In the simulated supply chain, the combined regression across tiers yields a positive coefficient (β = +109.6, p = 0.015), meaning that as one moves from factory toward retailer

What carries the argument

The Autonomy Score (AS) is the load-bearing measurement: it combines the Initiative Rate (IR), the fraction of executable tasks the agent generates internally rather than receiving as external delegation, with the Consultation Rate (CR), the fraction of tasks that require external approval before proceeding, as AS = w1·IR + w2·(1−CR). Because both rates are computed from observable task-store events—task creation, status updates, and consultation requests—the score can be logged continuously during runtime. The score is what turns the conceptual claim that autonomy is about how decisions are distributed, not how well they are made, into a number that can be tracked, compared across tiers, an

Load-bearing premise

The simulated user in the consultation mechanism always approves every request, so the decision-control component counts how often the agent asks for approval, not whether an external decision-maker ever blocks or changes an action; if real users vetoed or revised plans, the measured relationship between autonomy score and cost could shift materially.

What would settle it

Re-run the beer distribution game with a user model that rejects a fixed share of consultations, say 20%, or that edits order quantities before approval, and check whether the combined tier-slope coefficient (β = +109.6) survives. If the sign or significance changes when rejections are possible, the positional autonomy-cost claim is an artifact of the always-approve user; if it persists, the claim is robust to realistic oversight.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Continuous runtime monitoring of AS would let an organization detect autonomy drift in an agentic system before it causes operational harm.
  • Governance policies can be expressed as tolerated ranges of AS per task or per tier, turning abstract human-oversight requirements into measurable thresholds.
  • The positional result suggests deployment guidance: grant higher autonomy to upstream planning and coordination, but keep tighter oversight in downstream, demand-facing operations.
  • Because AS is implementation-independent, the same metric could be used to compare autonomy across heterogeneous agentic platforms and across lifecycle stages.
  • The finding that autonomy does not improve performance in a uniform way warns against designing agentic systems with maximum autonomy as the objective.

Where Pith is reading between the lines

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

  • A sharper test of the claim that autonomy is orthogonal to capability would hold the underlying model fixed and vary only the interaction policy, so that any cost differences cannot be attributed to the agent's reasoning quality; the current prompt-based manipulation may blend the two.
  • The upstream/downstream split may reflect bullwhip-signal dynamics: upstream tiers face delayed, amplified demand signals where autonomous planning smooths decisions, while retailers see true demand and autonomy amplifies overreaction. This could be tested by regressing tier-level bullwhip ratios on AS.
  • The framework's collaboration dimension is left out of the score; a natural extension is to assign partial ownership weights to shared tasks, which would let multi-agent collaboration be scored without collapsing into delegation or consultation.
  • If real consultations were allowed to reject or modify plans, the decision-control component would reflect genuine human veto power, and the observed autonomy-cost relationship could change; the paper's own always-approve user makes the current estimate a measure of how often the agent asks, not whether it is actually controlled.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes the Agentic AI Autonomy Assessment (AAAA) framework, which operationalizes autonomy as a continuous, task-level construct using three interaction mechanisms: delegation, consultation, and collaboration. Two are quantified via an Initiative Rate (IR, Eq. 1) and a Consultation Rate (CR, Eq. 2), combined into an Autonomy Score AS = w1·IR + w2·(1−CR) (Eq. 3). The framework is applied to a simulated beer distribution game with LLM-based agents at four supply chain tiers under five prompt-level autonomy configurations (L1–L5). The central empirical claim is a tier-dependent relationship between AS and tier costs: upstream tiers (factory, distributor) benefit from higher autonomy, while downstream tiers (retailer) are harmed, with a pooled coefficient β=+109.6 (p=0.015) for the slope trend across tiers. The paper interprets this as evidence that autonomy is orthogonal to capability and should be governed as a context-dependent design variable.

Significance. If the construct validity of AAAA were established, the framework would address a real gap in agentic AI governance: existing autonomy taxonomies are discrete, subjective, and not lifecycle-monitorable. The paper's strengths are its concrete, implementation-independent task-interaction operationalization; the use of a well-understood simulation testbed; the transparent acknowledgment of several limitations (prompt-based manipulation, narrow AS range, simulation context, excluded collaboration dimension); and the articulation of a falsifiable positional prediction. However, the current empirical support is undermined by three load-bearing problems: (i) the weights in Eq. (3) are never specified, so all reported coefficients are functions of an unreported measurement choice; (ii) the consultation user is an always-approve function, so the decision-control component measures consultation frequency, not actual external control; and (iii) the statistical evidence for the positional effect is thin and uncorrected for multiple testing. The orthogonality-to-capability claim is a design axiom, not an empirical result. These issues are fixable within the manuscript's scope, but the paper's c

major comments (4)
  1. [§3.4, Eq. (3); §4.3, Table 7] The weights w1 and w2 in Eq. (3) are defined only by the constraints w1,w2 ≥ 0 and w1+w2 = 1, but the values used in the experiments are never reported. Every AS value in the 240 runs, and therefore every coefficient in Table 7 and the pooled β=+109.6, depends on this unreported choice. With w1=1 the score ignores consultation entirely; with w2=1 it reduces to 1−CR and ignores initiative. These extremes can rank runs differently, and the paper gives no sensitivity analysis. The headline empirical result is therefore not reproducible or interpretable until the weights are specified and robustness across the weight space is demonstrated.
  2. [§4.2.1; §3.4, Eq. (2)] The consultation mechanism is implemented with a user model that "always returns approval" and is described as having no influence on model actions. Under this design, CR counts how often the agent asks, not whether an external decision-maker actually controls or changes the action. The decision-control dimension of AS therefore has low construct validity: an agent that asks frequently but is never vetoed or redirected is scored as less autonomous even though its actual control over outcomes is unchanged. The reported relationship between AS and cost could differ materially under a user who can reject or modify consultations. The authors should either implement a realistic user model or explicitly reframe CR as "consultation frequency" and soften claims about external oversight.
  3. [§4.3, Table 7, Figs. 6–7] The statistical basis for the positional effect is fragile. Only the distributor (p=0.028) and retailer (p=0.018) cost coefficients are significant at the 0.05 level, with no multiple-comparison correction; factory (p=0.167) and wholesaler (p=0.573) are not. For fill rate, only the retailer is significant (p=0.030). The pooled regression producing β=+109.6 (p=0.015) is described only narratively; the estimation method is not specified, and with only four tier-level observations the p-value is highly dependent on the model. The authors should report the exact pooled model, account for the uncertainty in tier-level slope estimates, present confidence intervals, and state whether the conclusions survive correction for the multiple tier-level tests.
  4. [Abstract, §5, §6] The claim that autonomy is "orthogonal to system capability" is not empirically established by this study. AS is constructed exclusively from initiative and consultation measures (Eq. 3) and contains no capability or performance term. The experiment does not measure capability separately; it measures cost and fill rate, which are outcomes that depend on many factors. Thus the orthogonality thesis is a consequence of the metric's definition, not a finding of the simulation. To support this claim, the paper would need to correlate AS with an independent measure of decision quality or capability, or at least explicitly state that orthogonality is an assumption of the framework rather than a demonstrated result.
minor comments (6)
  1. [§4.2.2] The text says "For each tier, the experiment is repeated twelve times" and then reports 60 runs per tier and 240 total. Clarify whether the twelve repetitions are per autonomy level (5 levels × 12 = 60) or per tier in total; the current wording is ambiguous.
  2. [Table 7] The Tobit coefficients and p-values for fill rate are not explained. Why are the factory Tobit entries missing ("-")? Also, clarify why Tobit is used for fill rate but not for cost, and define the censoring mechanism.
  3. [Fig. 5 caption] Typo: "Implmented" should be "Implemented".
  4. [§3.3, §3.4] Collaboration is introduced as one of three interaction mechanisms but is excluded from the quantitative AS. The paper calls this extensible, but the abstract and introduction present three dimensions; the metric operationalizes only two. This should be stated more prominently so readers are not misled.
  5. [Table 5] The demand pattern is deterministic and known in advance. This limits the scope of the “uncertainty” narrative for the bullwhip effect; a brief note on this, or a stochastic-demand extension, would strengthen the experimental realism.
  6. [§3.4, Eqs. (1)–(2)] The denominator N is the total number of executable tasks, but agents can create and remove tasks via create_task and remove_task. This makes N agent-controlled and could distort IR and CR. A short discussion of this endogeneity and its potential impact on the metrics would be useful.

Circularity Check

1 steps flagged

Headline 'orthogonal to capability' claim is built into the AS definition, while the empirical AS–cost regressions are not circular.

specific steps
  1. self definitional [§3.4, Eq. (3); §6 Conclusions]
    "AS=w 1IR+w 2(1−CR) ... It is important to emphasise that the proposed Autonomy Score should not be interpreted as a measure of intelligence, reasoning quality, or decision performance. ... the results reinforce the notion that autonomy is orthogonal to system capability, capturing how decisions are made rather than how well they are made."

    The 'orthogonal to capability' conclusion is entailed by the operationalisation rather than discovered by the experiment. Eq. (3) defines AS using only Initiative Rate and (1 − Consultation Rate), and the same section states that AS 'should not be interpreted as a measure of intelligence, reasoning quality, or decision performance.' Capability terms are absent by construction, so claiming that autonomy is orthogonal to capability is a restatement of the metric definition. The AS–cost regressions add an independent empirical association, but the headline orthogonality claim was already contained in the input definition.

full rationale

The only defensible circularity is the self-definitional framing of the Autonomy Score as 'orthogonal to capability': AS is defined exclusively from initiative and consultation counts, so the claim that it does not measure capability is true by construction rather than being a new result. The central empirical finding—the position-dependent AS–cost relationship—is not circular: AS is a count of task initiations and consultations, while costs come from the LangGraph beer-game simulation, so the pooled β=+109.6 is not algebraically forced by Eq. (3). The unreported weights w1,w2 in Eq. (3) are a serious reproducibility and interpretability defect, but they are not a circular reduction because the paper does not state that they were fitted to the outcome. The §6 limitations (prompt-based autonomy manipulation, narrow AS range) and §4.2.1's always-approve consultation model undermine construct validity and external validity, but they are not cases where a fitted input is renamed a prediction. The repeated Hendriksen (2023) self-citation is background motivation, not load-bearing for the metric or regressions. Overall, the measurement-plus-regression core retains independent empirical content; the circularity is confined to the definitional headline claim about orthogonality to capability.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 1 invented entities

The AAAA framework is definitional: autonomy is equated with task-initiation and non-consultation rates. The empirical experiment then depends on prompt-level assumptions, an always-approving user, and the beer-game context. No external benchmark, code, or data is supplied to independently validate the construct.

free parameters (2)
  • w1, w2 (AS weights) = not reported
    The Autonomy Score in Eq. (3) depends on weighting coefficients w1 and w2, but the values used in the experiment are never stated. The regression results are therefore conditional on an unreported weighting choice.
  • L1-L5 system-prompt augmentation conditions = five hand-written prompt variants
    Chosen by the authors to induce different autonomy behaviours; no manipulation check connects them to distinct AS values, and the observed AS distribution is narrow and skewed.
axioms (5)
  • domain assumption Task-level interactions (delegation, consultation, collaboration) are a sufficient basis for measuring agentic autonomy.
    Sections 3.1-3.3 postulate this; no proof is offered that these three interaction types exhaust the autonomy-relevant events.
  • domain assumption Consultations can be counted objectively from tool calls and represent decision control.
    Section 3.4 and 4.2.1 operationalise CR as the rate of consultation requests; in the experiment the user always approves, so CR measures only the request rate, not actual external control.
  • ad hoc to paper Prompt-level manipulation L1-L5 changes agentic autonomy in the intended direction.
    Section 4.2.2; the observed skew and narrow AS range suggest limited control, and no per-level AS comparison or manipulation check is reported.
  • domain assumption The beer distribution game's simulated costs and fill rates are a valid proxy for supply chain performance.
    Section 4.1 uses a well-known game, but the 10-week horizon and synthetic seasonal demand are simplifications that may not generalise.
  • standard math OLS and tobit regression assumptions hold for these data.
    Section 4.3 reports OLS and tobit coefficients without residual diagnostics, robust standard errors, or variance checks.
invented entities (1)
  • Autonomy Score (AS) no independent evidence
    purpose: Composite metric to quantify autonomy from initiative rate and consultation rate.
    AS is a definitional construct; no external benchmark or inter-rater validation shows it tracks autonomy independently of the framework's own definitions.

pith-pipeline@v1.3.0-alltime-deepseek · 19317 in / 14392 out tokens · 145490 ms · 2026-08-01T02:30:59.853787+00:00 · methodology

0 comments
read the original abstract

Supply chain decision-making is rapidly transforming with the rise of agentic AI - highly autonomous systems that can operate on complex, long-horizon tasks. Yet the adoption of agentic systems outpaces their governance: existing taxonomies of autonomy only offer discrete classifications, rely on subjective judgement, and cannot track autonomy across a system's life cycle, leaving enterprises unable to assess the risks of increasingly autonomous supply chain agents. This paper proposes the Agentic AI Autonomy Assessment (AAAA) framework, which defines and measures the degree of autonomy at a task level. The framework is based on the three dimensions of user delegation, consultation, and collaboration, enabling continuous monitoring from an agent's development through its runtime to end-of-life. The framework's construct validity was tested in a simulated beer distribution game, examining how the autonomy score relates to a company's performance. Results reveal a weak link between autonomy and tier costs with a positional effect: upstream tiers benefit from higher autonomy while downstream tiers are harmed, positioning autonomy as an inherent dimension of agentic systems, orthogonal to capability. The framework provides a foundation for risk assessment, governance, and transparent autonomy policies to support the governed enterprise adoption of agentic AI in supply chains.

Figures

Figures reproduced from arXiv: 2607.25405 by Christian Hendriksen, Elias Ribeiro da Silva, Lennart Trumpler, Luis Antonio de Santa-Eulalia, Rodrigo Furlan de Assis.

Figure 1
Figure 1. Figure 1: Distribution of human, shared, and agent responsibility across stages of the goal realization process under different autonomy configurations. The horizontal marker indicates the stage at which the agent assumes primary decision responsibility. Adapted from autonomy-level frameworks proposed in Feng et al. (2025); Kasirzadeh and Gabriel (2025); Morris et al. (2023) [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: presents the conceptual organisation of the proposed framework. The frame￾work is organised around three interacting components: the agentic system, external decision-making entities, and the operational environment. Together, these compo￾nents form a continuous perception-reasoning-action loop in which goals are generated, decomposed into executable tasks, executed, and continuously updated according to c… view at source ↗
Figure 3
Figure 3. Figure 3: Task interaction mechanisms considered by the proposed framework. The three interaction mech￾anisms represent alternative distributions of responsibility over the same executable task. Solid arrows denote the primary execution flow, whereas dashed arrows represent information exchange and coordination interac￾tions. The three interaction mechanisms illustrated in [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Beer Distribution Supply Chain The parameters for the simulation are described in [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Agentic Workflow Graph Implmented using LangGraph manage the simulation workflow (setting up a new game, advancing the week, querying state) and endpoints for company interactions (creating shipments, placing orders). The agentic workflow is implemented as a graph, consisting of a main game graph that controls the overall simulation flow and separate sub-graphs for each tier [PITH_FULL_IMAGE:figures/full_… view at source ↗
Figure 6
Figure 6. Figure 6: Autonomy-Cost Relationship Across Supply Chain Tiers Indicating a Performance Split Between Upstream and Downstream Actors To support further analysis of this pattern, the slopes of each tier are compared and combined into a pooled linear regression model, based on all collected data points. The left-hand side of [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Combined Linear Regression Model for the Relation of Supply Chain Tier with Simulation Cost (2004), which argues that autonomy primarily influences how decision-making respon￾sibilities are distributed, rather than directly determining performance outcomes. This distinction is critical, as it positions autonomy as a governance variable rather than a performance variable (Chiris & Mishra, 2025). In this sen… view at source ↗

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

103 extracted references · 13 canonical work pages

  1. [1]

    and Chatterjee, Samir , month = dec, year =

    Peffers, Ken and Tuunanen, Tuure and Rothenberger, Marcus A. and Chatterjee, Samir , month = dec, year =. A. Journal of Management Information Systems , publisher =. doi:10.2753/MIS0742-1222240302 , abstract =

    doi:10.2753/mis0742-1222240302

  2. [2]

    , year =

    Sterman, John D. , year =. Business dynamics: systems thinking and modeling for a complex world , isbn =

  3. [3]

    , year =

    Montgomery, Douglas C. , year =. Design and analysis of experiments , isbn =

  4. [4]

    Feng, K. J. Kevin and McDonald, David W. and Zhang, Amy X. , month = jul, year =. Levels of. doi:10.48550/arXiv.2506.12469 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2506.12469

  5. [5]

    Evaluation of the

    Bandi, Ajay and Zeng, Ruida , editor =. Evaluation of the. Computer. 2025 , keywords =. doi:10.1007/978-3-031-76273-4_5 , abstract =

    work page doi:10.1007/978-3-031-76273-4_5 2025

  6. [6]

    Dalkey, Norman and Helmer, Olaf , month = apr, year =. An. Management Science , publisher =. doi:10.1287/mnsc.9.3.458 , abstract =

    work page doi:10.1287/mnsc.9.3.458

  7. [7]

    , year =

    Yin, Robert K. , year =. Case study research and applications: design and methods , isbn =

  8. [8]

    Optimizing delegation between human and

    Fuchs, Andrew and Passarella, Andrea and Conti, Marco , booktitle=. Optimizing delegation between human and. 2023 , organization=

    2023

  9. [9]

    IEEE Access , author =

    Agentic. IEEE Access , author =. 2025 , keywords =. doi:10.1109/ACCESS.2025.3532853 , abstract =

    arXiv 2025

  10. [10]

    Future Internet , author =

    The. Future Internet , author =. 2025 , keywords =. doi:10.3390/fi17090404 , abstract =

    doi:10.3390/fi17090404 2025

  11. [11]

    International Journal of Production Economics , author =

    Will bots take over the supply chain?. International Journal of Production Economics , author =. 2021 , keywords =. doi:10.1016/j.ijpe.2021.108279 , abstract =

    arXiv 2021

  12. [12]

    , month = jul, year =

    Vaccaro, Michelle and Caosun, Michael and Ju, Harang and Aral, Sinan and Curhan, Jared R. , month = jul, year =. Advancing. doi:10.48550/arXiv.2503.06416 , abstract =

    doi:10.48550/arxiv.2503.06416

  13. [13]

    Pactum , author =

    The leader in agentic. Pactum , author =

  14. [14]

    Jannelli, Valeria and Schöpf, Stefan and Bickel, Matthias and Netland, Torbjørn and Brintrup, Alexandra , month = dec, year =. Agentic. International Journal of Production Research , publisher =. doi:10.1080/00207543.2025.2604311 , abstract =

    arXiv 2025

  15. [15]

    From natural language to simulations: applying

    Jackson, Ilya and Jesus Saenz, Maria and Ivanov, Dmitry , month = feb, year =. From natural language to simulations: applying. International Journal of Production Research , publisher =. doi:10.1080/00207543.2023.2276811 , abstract =

    arXiv 2023

  16. [16]

    Alignment,

    Tallam, Krti , month = feb, year =. Alignment,. doi:10.48550/arXiv.2503.05748 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.05748

  17. [17]

    and Wiest, Olaf and Zhang, Xiangliang , month = apr, year =

    Guo, Taicheng and Chen, Xiuying and Wang, Yaqi and Chang, Ruidi and Pei, Shichao and Chawla, Nitesh V. and Wiest, Olaf and Zhang, Xiangliang , month = apr, year =. Large. doi:10.48550/arXiv.2402.01680 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.01680

  18. [18]

    2026 , keywords =

    Emerging. 2026 , keywords =. doi:10.1007/978-3-032-01218-0 , language =

    work page doi:10.1007/978-3-032-01218-0 2026

  19. [19]

    , year =

    Sterman, John D. , year =. Modeling. Management Science , publisher =

  20. [20]

    Sloan Management Review , author =

    The. Sloan Management Review , author =

  21. [21]

    doi:10.48550/arXiv.2210.03629 , abstract =

    Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , month = mar, year =. doi:10.48550/arXiv.2210.03629 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2210.03629

  22. [22]

    doi:10.48550/arXiv.2511.02424 , abstract =

    Choi, Jae-Woo and Kim, Hyungmin and Ong, Hyobin and Yoon, Youngwoo and Jang, Minsu and Kim, Dohyung and Kim, Jaehong , month = feb, year =. doi:10.48550/arXiv.2511.02424 , abstract =

    doi:10.48550/arxiv.2511.02424

  23. [23]

    arXiv.org , author =

    Dive into. arXiv.org , author =. doi:10.48550/arXiv.2604.14228 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2604.14228

  24. [24]

    Computers & Industrial Engineering , volume=

    Resilient food supply chain design: Modelling framework and metaheuristic solution approach , author=. Computers & Industrial Engineering , volume=. 2019 , publisher=. doi:10.1016/j.cie.2019.05.011 , pages=

    work page doi:10.1016/j.cie.2019.05.011 2019

  25. [25]

    The role of agentic

    Hosseini, Soodeh and Seilani, Hossein , journal=. The role of agentic. 2025 , publisher=

    2025

  26. [26]

    Computers in Industry , volume=

    On implementing autonomous supply chains: A multi-agent system approach , author=. Computers in Industry , volume=. 2024 , publisher=

    2024

  27. [27]

    The impact of digital technology and

    Ivanov, Dmitry and Dolgui, Alexandre and Sokolov, Boris , journal=. The impact of digital technology and. 2019 , publisher=

    2019

  28. [28]

    Journal of manufacturing systems , volume=

    Knowledge-based multi-agent system for manufacturing problem solving process in production plants , author=. Journal of manufacturing systems , volume=. 2018 , publisher=

    2018

  29. [29]

    Alva, Linga Reddy and Pandey, Bishwajeet , journal=. Agentic. 2026 , publisher=

    2026

  30. [30]

    Drivers and constraints of agentic

    Trifone, Nicol. Drivers and constraints of agentic. International Journal of Production Research , pages=. 2026 , publisher=

    2026

  31. [31]

    Agentic digital twins: Bridging model-based and

    Ivanov, Dmitry , journal=. Agentic digital twins: Bridging model-based and. 2026 , publisher=

    2026

  32. [32]

    2026 , publisher=

    Ahi, Kiarash and Agrawal, Vaibhav and Valizadeh, Saeed , journal=. 2026 , publisher=

    2026

  33. [33]

    and Li, Keyao and Appanderanda, Mandanna and Al-Bashrawi, Mousa Ahmad and Chae, Inyoung , month = jul, year =

    Hughes, Laurie and Dwivedi, Yogesh K. and Li, Keyao and Appanderanda, Mandanna and Al-Bashrawi, Mousa Ahmad and Chae, Inyoung , month = jul, year =. Journal of Global Information Technology Management , publisher =. doi:10.1080/1097198X.2025.2524286 , abstract =

    arXiv 2025

  34. [34]

    2025 , publisher=

    Ren, Yinwang and Liu, Yangyang and Ji, Tang and Xu, Xun , journal=. 2025 , publisher=

    2025

  35. [35]

    Nisa, Ume and Shirazi, Muhammad and Saip, Mohamed Ali and Pozi, Muhammad Syafiq Mohd , journal=. Agentic. 2025 , publisher=

    2025

  36. [36]

    The autonomy equation: How agentic

    Cossatin, Angelo Geninatti and Ferrero, Fabio and Ardissono, Liliana and Mauro, Noemi , journal=. The autonomy equation: How agentic. 2026 , publisher=

    2026

  37. [37]

    Omega , pages=

    Supply chain digital twin design and implementation at scale: A case study of Ford Motors Company and generalizations , author=. Omega , pages=. 2025 , publisher=

    2025

  38. [38]

    Generative

    El Baz, Hicham and Ghali, Mohammed-Khalil and Andrus, Linda and Won, Daehan and Yoon, Sang Won and Wang, Yong , journal=. Generative. 2026 , publisher=

    2026

  39. [39]

    The Autonomous Enterprise: How Agentic

    Kshetri, Nir , journal=. The Autonomous Enterprise: How Agentic. 2026 , publisher=

    2026

  40. [40]

    Journal of Modelling in Management , pages=

    Agentic automation driven reinforcement learning for inventory optimization , author=. Journal of Modelling in Management , pages=. 2026 , publisher=

    2026

  41. [41]

    Tiwari, Nitin and Prasad, Subrat Kumar , booktitle=. Agentic. 2025 , organization=

    2025

  42. [42]

    EcoptiAI: E-commerce process optimization and operational cost minimization through task automation using agentic

    Alecsoiu, Olivia-Roxana and Faruqui, Nuruzzaman and Panagoret, Andreea Anamaria and Ionu. EcoptiAI: E-commerce process optimization and operational cost minimization through task automation using agentic. IEEE Access , volume=. 2025 , publisher=

    2025

  43. [43]

    IEEE Transactions on Engineering Management , volume=

    Integrating Supply chain viability and regeneration toward survival of ecosystems , author=. IEEE Transactions on Engineering Management , volume=. 2026 , publisher=

    2026

  44. [44]

    Nature Computational Science , volume=

    The evolution of digital twins from reactive to agentic systems , author=. Nature Computational Science , volume=. 2026 , publisher=

    2026

  45. [45]

    Integrating agentic

    Hasan, Agus and Nguyen, Dong Trong , journal=. Integrating agentic. 2026 , publisher=

    2026

  46. [46]

    The rise of agentic

    Islam, Md Asadul and Somu, Subbulakshmi and Aldaihani, Faraj Mazyed Faraj , journal=. The rise of agentic. 2026 , publisher=

    2026

  47. [47]

    Engineering applications of artificial intelligence , volume=

    A comprehensive review of simulation optimization methods in agricultural supply chains and transition towards an agent-based intelligent digital framework for agriculture 4.0 , author=. Engineering applications of artificial intelligence , volume=. 2025 , publisher=

    2025

  48. [48]

    Khapre, Smita and Mersha, Melkamu Abay and Olalekan, Zainab and Shakil, Hassan and Iskander, Samer and Moustafa, Asser and Kalita, Jugal , journal=. Agentic. 2026 , publisher=

    2026

  49. [49]

    Supply Chain with Sixth Sense Agentic

    Parthasarathy, Venkatesh Prabu and Vadigicherla, Madhusudan Sharma , booktitle=. Supply Chain with Sixth Sense Agentic. 2026 , organization=

    2026

  50. [50]

    Journal of Systems and Information Technology , pages=

    Dynamics of human--agentic information systems relationship: a systematic literature review , author=. Journal of Systems and Information Technology , pages=. 2026 , publisher=

    2026

  51. [51]

    Information & Management , pages=

    Agentic Artificial Intelligence as a New Frontier in Information Systems: Promise, Peril, and Research Opportunities , author=. Information & Management , pages=. 2026 , publisher=

    2026

  52. [52]

    Trust in

    Afroogh, Saleh and Akbari, Ali and Malone, Emmie and Kargar, Mohammadali and Alambeigi, Hananeh , journal=. Trust in. 2024 , publisher=

    2024

  53. [53]

    Al-Sharafi, Mohammed A and Al-Zaeemi, Shehab Abdulhabib and Dwivedi, Yogesh K and Abdulkareem, Zainab O and Ali, Ahmed and Chae, Inyoung , journal=. Agentic. 2026 , publisher=

    2026

  54. [54]

    Research framework and evolutionary trends of smart supply chain in

    Zhang, Jingrong and Hou, Yanning and Song, We and Shen, Jingwen and Li, Yumin , journal=. Research framework and evolutionary trends of smart supply chain in. 2026 , publisher=

    2026

  55. [55]

    International Journal of Production Research , pages=

    Large language models in supply chain management: a systematic literature review and application framework , author=. International Journal of Production Research , pages=. 2026 , publisher=

    2026

  56. [56]

    2025 , publisher=

    Sapkota, Ranjan and Roumeliotis, Konstantinos I and Karkee, Manoj , journal=. 2025 , publisher=

    2025

  57. [57]

    Levels of

    Morris, Meredith Ringel and Sohl-Dickstein, Jascha and Fiedel, Noah and Warkentin, Tris and Dafoe, Allan and Faust, Aleksandra and Farabet, Clement and Legg, Shane , journal=. Levels of

  58. [58]

    Homo agenticus in the age of agentic

    Leonardi, Paul M , journal=. Homo agenticus in the age of agentic. 2025 , publisher=

    2025

  59. [59]

    M. Agentic. Expert Systems with Applications , pages=. 2026 , publisher=

    2026

  60. [60]

    International Journal of Production Economics , volume=

    Will bots take over the supply chain? Revisiting agent-based supply chain automation , author=. International Journal of Production Economics , volume=. 2021 , publisher=

    2021

  61. [61]

    Journal of Industrial Information Integration , volume=

    Towards autonomous supply chains: Definition, characteristics, conceptual framework, and autonomy levels , author=. Journal of Industrial Information Integration , volume=. 2024 , publisher=

    2024

  62. [62]

    2026 , publisher=

    Hasselwander, Marc , journal=. 2026 , publisher=

    2026

  63. [63]

    Narayana, Shivaprasad Sankesha , booktitle=. Agentic. 2025 , organization=

    2025

  64. [64]

    Abou Ali, Mohamad and Dornaika, Fadi and Charafeddine, Jinan , journal=. Agentic. 2025 , publisher=

    2025

  65. [65]

    Thought Management System for long-horizon, goal-driven

    Bidochko, Andrii and Vyklyuk, Yaroslav , journal=. Thought Management System for long-horizon, goal-driven. 2025 , publisher=

    2025

  66. [66]

    arXiv preprint arXiv:2210.02406 , year=

    Decomposed prompting: A modular approach for solving complex tasks , author=. arXiv preprint arXiv:2210.02406 , year=

    Pith/arXiv arXiv

  67. [67]

    Findings of the Association for Computational Linguistics: NAACL 2024 , pages=

    Adapt: As-needed decomposition and planning with language models , author=. Findings of the Association for Computational Linguistics: NAACL 2024 , pages=

    2024

  68. [68]

    Memetic Computing , volume=

    A novel multi-agent architecture based on decomposition and learning automata to hybridize multi-objective metaheuristics , author=. Memetic Computing , volume=. 2025 , publisher=

    2025

  69. [69]

    IEEE Transactions on Audio, Speech and Language Processing , year=

    Chain-of-Cognition Through Multi-Agent Collaboration Against Cross-Modal Cognitive Interference , author=. IEEE Transactions on Audio, Speech and Language Processing , year=

  70. [70]

    arXiv preprint arXiv:2310.10134 , year=

    Clin: A continually learning language agent for rapid task adaptation and generalization , author=. arXiv preprint arXiv:2310.10134 , year=

    Pith/arXiv arXiv

  71. [71]

    Architectures for

    Nowaczyk, Sławomir , month = dec, year =. Architectures for. doi:10.48550/arXiv.2512.09458 , abstract =

    doi:10.48550/arxiv.2512.09458

  72. [72]

    Navigating autonomy and control in human-

    Adam, Martin and Diebel, Christopher and Goutier, Marc and Benlian, Alexander , journal=. Navigating autonomy and control in human-. 2024 , publisher=

    2024

  73. [73]

    Decision science in the agentic era: Research frontiers on agentic

    Zhang, Yingjie and Sun, Tianshu , journal=. Decision science in the agentic era: Research frontiers on agentic. 2026 , publisher=

    2026

  74. [74]

    The rise of agentic

    Murugesan, San , journal=. The rise of agentic. 2025 , publisher=

    2025

  75. [75]

    Toward the Internet of Agentic

    Ren, Yuzheng and Yang, Jiahui and Zhang, Haijun and Yu, Fei Richard and Xiao, Yifu and Cao, Xueyan and Sun, Chunlei and He, Ying , journal=. Toward the Internet of Agentic. 2026 , publisher=

    2026

  76. [76]

    Software Architecture in the Age of Agentic

    Vaidhyanathan, Karthik and Muccini, Henry , booktitle=. Software Architecture in the Age of Agentic. 2025 , organization=

    2025

  77. [77]

    Cognition , volume=

    Attribution of intentional agency towards robots reduces one’s own sense of agency , author=. Cognition , volume=. 2020 , publisher=

    2020

  78. [78]

    Sustainable Development , year=

    Toward sustainable agentic ai systems: A survey of architectures and methodologies , author=. Sustainable Development , year=

  79. [79]

    Characterizing

    Kasirzadeh, Atoosa and Gabriel, Iason , journal=. Characterizing

  80. [80]

    Humanities and Social Sciences Communications , publisher =

    Chen, Lin and Zhang, Yunke and Feng, Jie and Chai, Haoye and Zhang, Honglin and Fan, Bingbing and Ma, Yibo and Zhang, Shiyuan and Li, Nian and Liu, Tianhui and Sukiennik, Nicholas and Zhao, Keyu and Li, Yu and Liu, Ziyi and Xu, Fengli and Li, Yong , month = apr, year =. Humanities and Social Sciences Communications , publisher =. doi:10.1057/s41599-026-07...

    work page doi:10.1057/s41599-026-07316-7

Showing first 80 references.