Enabling Supply Chain Resilience Through Green Supply Chain Integration in Agribusinesses: The Role of Responsible Innovation and Agentic AI

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a very interesting paper exploring the relationship between green supply chain integration and supply chain resilience. The manuscript could be suitable for publication but there are a few issues that the authors must address.

GSCI is operationalised as a single construct combining internal employee integration with supplier and customer integration. However, much of the theoretical argument concerns the complexity and information-processing demands at the supply chain level. Please justify aggregating these dimensions and provide evidence that a higher-order GSCI construct is appropriate. A robustness analysis separating internal and external integration, and ideally supplier and customer integration, would help establish whether the reported non-linearity is genuinely driven by cross-organisational integration.

The information processing explanation for the inverted U-shape remains inferential because neither information overload, information-processing demand, information-processing capacity nor coordination cost is measured. The discussion also appears to combine information volume, information quality, ambiguity and resource cost. Please distinguish these mechanisms more clearly, consider plausible alternative explanations for the quadratic relationship, and moderate causal claims about the mechanism unless additional evidence or robustness tests can be supplied.

The sampling and screening process for agentic AI requires more transparency. Please report the dates of each wave, the number of firms entering and being excluded at each screening and survey stage, and the characteristics of firms excluded by the agentic-AI screen. Because the final sample appears to contain only firms claiming relatively advanced autonomous AI functionality, a discussion of how these functionalities were verified is needed. To what extent is the survey measuring Agentic AI capabilities or advanced automation/analytics?

Given that all principal constructs are latent multi-item variables and the model includes mediation, non-linearity and moderation, please explain the decision to estimate the structural model using composite-score hierarchical regression after conducting CFA. Please discuss the implications of measurement error and provide an appropriate robustness analysis, potentially using a latent-variable structural model or alternative factor-score specifications.

The manuscript frames the study as resolving a green–resilience tension and repeatedly claims a win-win outcome between greenness and resilience. However, environmental performance or realised greenness is not included in the empirical model; GSCI is an integration practice rather than an environmental outcome.

The interpretation of H5a and H5b appears stronger than the statistical evidence permits. In both models, GSCI² × agentic AI is non-significant, while only GSCI × agentic AI is significant. This supports a change in slope or turning-point location, not moderation of the quadratic curvature itself.

 

Minor issues:

• Table 5 marks the GSCI × AAI coefficient for SCR as 0.337***, but the text reports p < 0.100.
• Table 4 marks GSCI in Model 5 as 0.138+, whereas the text reports p < 0.010.
• ΔR² for Model 7 is printed as 0126.
• RMSEA for the one-factor model is reported as 1.178, which may be a typographical error for 0.178.
• AAI is labelled “AAC” in Table 3.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Resolving the Green-Resilience Tension in Agricultural Supply Chains: The Role of Responsible Innovation and Agentic AI

The article examines the effects of green supply chain integration (GSCI) on responsible innovation (RI) and supply chain resilience (SCR) in Chinese agri-food firms. The authors hypothesize that the GSCI–RI and GSCI–SCR relationships are inverted U-shaped. Responsible innovation is proposed as a mediator, while agentic AI is treated as a moderator of these relationships.

The study was conducted in three waves at six-month intervals. GSCI and agentic AI were measured in the first wave, responsible innovation in the second, and supply chain resilience in the third. The final sample comprises 225 firms. Hierarchical regression and the MEDCURVE procedure were used.

The findings indicate inverted U-shaped relationships between GSCI and both RI and SCR. The indirect effect through RI is positive at low levels of GSCI and negative at high levels. Agentic AI significantly moderates the linear component of both relationships, but not their curvature.

Overall assessment

The article addresses a timely and potentially original research problem by combining green supply chain integration, responsible innovation, supply chain resilience, and agentic AI. Its strengths include a clearly defined theoretical model, three-wave data collection, and an attempt to examine nonlinear relationships.

However, the manuscript requires substantial methodological and interpretive revisions. The main concerns are as follows:

• The organization of the three-wave study is not fully described in Section 3.1. The authors report six-month intervals and the allocation of constructs across waves, but do not provide the survey dates, response numbers at each stage, respondent attrition, procedures for matching questionnaires, or information on whether the same individual responded in all three waves. The sample flow and an attrition analysis should be reported.
• The sample includes only firms classified as advanced users of agentic AI (Section 3.1). The authors should report the number of firms excluded through the screening questions and provide the range and distribution of the AAI variable. This selection may restrict the variability of the moderator and limit the generalizability of the findings.
• The manuscript does not provide sufficient evidence that the systems used by the surveyed firms genuinely meet the defining characteristics of agentic AI rather than representing advanced analytics, automation, or AI-enabled decision support. The screening procedure is described, but it relies on respondents’ declarations. The criteria concerning autonomy, planning, and independent action initiation should be justified more thoroughly.
• The reporting and interpretation of H5a and H5b are not fully consistent (Section 4.3, Table 5, and the abstract). The GSCI × AAI interactions are significant, whereas the GSCI² × AAI interactions are not. Agentic AI therefore modifies the slope of the relationships, but there is no evidence that it significantly changes their curvature. H5a and H5b should be reported as only partially supported. The abstract’s statement that agentic AI “positively moderates the inverted U-shaped effects” is too broad.
• The result for H5b is reported inconsistently. In Table 5, the coefficient for GSCI × AAI → SCR is β = 0.337 and marked with ***, whereas the text reports p < 0.100. This value should be verified.
• The causal interpretation of H4 and the overall model is too strong. The analysis indicates a nonlinear statistical indirect effect consistent with the proposed mediation mechanism, but it does not by itself establish causal direction or exclude reverse causality and omitted-variable effects. The causal language used in the discussion and conclusions should therefore be moderated.
• The regression and moderation results are incompletely reported. Tables 4 and 5 do not provide standard errors, confidence intervals, or exact p-values, which makes it difficult to assess H5a and H5b. For H4, the authors report 5,000 bootstrap replications and confidence intervals for effects at GSCI ±1 SD, but they should also specify the type of bootstrap confidence intervals and provide the remaining estimation details.
• The presentation of the results contains errors or typographical mistakes: RMSEA = 1.178, probably 0.178 (Section 4.1); ΔR² = 0126, with a missing decimal point (Table 4); and AAI is labelled as AAC (Table 3).
• The ethical information is incomplete. Section 3.1 mentions confidentiality and the intended use of the data, but the manuscript does not clearly report ethics committee approval, the basis for an exemption, or the respondents’ informed and voluntary consent.
• The reference in Section 3.2 is incorrect: the measurement items are presented in Table 2, not Table 1. Table 2 reports the 24 principal measurement items, so the absence of item wording is not a concern. However, the manuscript does not provide the complete survey instrument, including the screening and background questions and information about matching responses across waves.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

All of my comments have been addressed in detail and the authors have actually made further improvements to the paper.  

Author Response

We are deeply grateful for your encouraging and generous assessment of our revised work. It is a great honor to learn that our point-by-point responses have been well received and, more importantly, that the reviewer has recognized our efforts to go beyond the initial requests in further refining the paper. We truly appreciate your time, expertise, and constructive guidance throughout this process, which have significantly strengthened the quality of our manuscript. Should the reviewer or the editorial office have any additional suggestions, we remain more than happy to address them promptly.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have addressed most of my comments, and the manuscript has been substantially improved. However, three issues still require clarification.

First, the descriptive statistics for agentic AI in Section 3.1 (M = 5.720, SD = 0.819) differ from those in Table 3 (M = 5.350, SD = 0.945), although both refer to the final sample of 225 firms. The authors should verify which values are correct.

Second, the sample-flow description should distinguish between attrition and exclusions during data screening. A total of 155 respondents were lost between waves (114 after Wave 1 and 41 after Wave 2), while 32 completed Wave 3 questionnaires were subsequently excluded during quality control. Therefore, all 187 cases should not be described as respondents who “dropped out during Waves 2 and 3.” The authors should also clarify which cases were included in the attrition-bias analysis.

Finally, the calculation of the turning-point shifts reported in Section 4.4 should be explained in sufficient detail to make the results reproducible.

Author Response

Please see the attachment

Author Response File: Author Response.pdf