Pith. sign in

REVIEW 3 major objections 5 minor 38 references

Synchronization Dynamics of Heterogeneous, Collaborative Multi-Agent AI Systems

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a phase-amplitude Kuramoto model, in which an agent's task progress is a phase and its influence is an amplitude, captures how heterogeneous AI agents synchronize, so that stronger coupling robustly raises the…

desk verdict A clearly written analogy between Kuramoto dynamics and multi-agent AI, but the 'rigorous mathematical foundation' claim outruns the evidence; the mapping is asserted, not validated. read the letter →

arxiv 2508.12314 v1 pith:3RH4FBXW submitted 2025-08-17 cs.MA cs.AInlin.AO

Chiranjit Mitra This is my paper · ORCID
classification cs.MAcs.AInlin.AO MSC 34C1534D0605C82
keywords Kuramotomodelmulti-agentAIsynchronizationchain-of-thoughtpromptingorderparameterscale-freenetworksagenticheterogeneousagents
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that the coordinated progress of heterogeneous AI agents toward a shared task can be described by a Kuramoto-type oscillator model in which each agent has a phase (task progress) and an amplitude (influence or resource level). It claims that this mapping is more than an analogy: the model's order parameter R(t) is a usable measure of collective coordination, and simulations on all-to-all and deterministic scale-free networks show that increasing coupling strength drives the system toward high R even when agents have different natural frequencies. A sympathetic reader would care because, if the mapping holds, task orchestration—how many agents to instantiate, how to connect them, how strongly they should communicate—becomes a quantitative design problem governed by synchronization theory rather than heuristic trial and error. The paper also proposes a correspondence between Chain-of-Thought reasoning and synchronization, treating each reasoning step as an iterative update toward a coherent collective solution.

What carries the argument

The central object is the phase-amplitude Kuramoto model of Eq. (3). It couples each agent's phase to neighbours through r_j \sin(\theta_j-\theta_i), letting high-amplitude agents dominate the pull, and couples each amplitude to neighbours through r_j \cos(\theta_j-\theta_i) on top of the logistic-type growth r_i(\$\lambda$-$r_i^{2}$). The order parameter R(t), identical in form to the complex Kuramoto order parameter but weighted by amplitudes, is the diagnostic: R near 1 indexes coherent task completion, and the paper uses its time average \langle R\rangle as the response variable in sweeps over coupling strength \epsilon and frequency spread \$\sigma$. The machinery carries the argument by turning qualitative intuitions about collaboration—weighted influence, specialization, resource sharing, emergent leadership—into a small set of tunable parameters that can be explored numerically.

What would settle it

Run a real multi-agent LLM system on a divisible task with N heterogeneous agents, vary the allowed inter-agent communication frequency or token budget (\epsilon), and measure the coherence of agent outputs (for example, pairwise agreement with a reference solution or an R-statistic on progress traces). The paper's claim predicts a monotone rise of this coherence proxy with \epsilon on both all-to-all and scale-free interaction graphs; observing a flat or non-monotone response, or no dependence on topology, would falsify the claimed foundation.

Watch

Extended reading notes

Core claim

The central claim is that the phase-amplitude Kuramoto model of Eq. (3), with phase dynamics \dot{\$\theta$}_i = \omega_i + \frac{\epsilon}{N}\sum_j A_{ij} r_j \sin(\theta_j-\theta_i) and radial dynamics \dot{r}_i = r_i(\$\lambda$ - $r_i^{2}$) + \frac{\epsilon}{N}\sum_j A_{ij} r_j \cos(\theta_j-\theta_i), provides a quantitative foundation for designing multi-agent AI systems. In this model, the agent's phase \theta_i tracks its position along the task's chain of thought, the amplitude r_i tracks its influence or allocated resources, \omega_i encodes processing speed or persona, and \epsilon times the adjacency structure A_{ij} encodes communication. The simulations show that the order parameter R(t) = \left|\frac{1}{N}\sum_j r_j(t)$e^{{i\theta_j(t)}}$\right| rises with \epsilon for both all-to-all (N=10) and deterministic scale-free (N=81) topologies, despite heterogeneous \omega drawn from normal distributions, and that larger frequency spread \$\sigma$ requires stronger coupling to reach the same R. On the strength of this, the paper claims a unified, physics-informed basis for optimizing agent count, topology, and resource-sharing policy in agentic AI workflows.

Load-bearing premise

The load-bearing premise is that an AI agent's task progress really does evolve like an oscillator phase pulled by sinusoidal phase differences, and that its influence or resource level follows the radial equation; the paper posits this correspondence without calibrating it against measured agent logs, so if the mapping fails, the simulations describe abstract oscillator dynamics rather than AI collaboration.

Editorial extensions

If this is right

  • If the model is right, tuning the communication strength \epsilon in a deployed agent network becomes an explicit control knob for coordination: raising \epsilon should push the team toward synchronized task completion.
  • Network topology becomes a design choice with predictable trade-offs: all-to-all connectivity gives rapid consensus, while a deterministic scale-free topology supports hierarchical structures typical of corporate agent workflows.
  • Heterogeneity among agents does not prevent synchronization; it only raises the coupling strength needed, so teams of specialized, differently paced agents can still converge if communication is strong enough.
  • The order parameter R(t) can serve as a runtime health metric for multi-agent systems: a drop in R signals loss of coordination, and orchestration logic could react by increasing coupling or rewiring the network.
  • Because amplitude r_i is interpreted as resource or token budget, the model implies that resource-sharing policy can be optimized to keep R high without starving any sub-task.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: a direct testable extension is to fit Eq. (3) to logs from a real LLM agent team, mapping message frequency or token budget to \epsilon and logged output agreement to R; the paper's claim predicts a monotone rise of R with \epsilon.
  • Beyond the paper: the Chain-of-Thought correspondence implies that internal representations of agents solving the same sub-task should converge stepwise during reasoning, a prediction that could be checked with representation-similarity measures across model layers.
  • Beyond the paper: adding an explicit orchestrator as a weak forcing or pinning term, rather than relying only on pairwise coupling, would connect this model to control-theoretic results on synchronization and could yield more robust convergence guarantees.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript proposes a phase-amplitude extension of the Kuramoto model, Eq. (3), as a model for heterogeneous multi-agent AI systems, with agent phase interpreted as task progress, amplitude as influence or resources, and coupling as communication. It defines an amplitude-weighted order parameter R(t) in Eq. (4), draws a qualitative correspondence between Chain-of-Thought prompting and synchronization in Section I.A, and reports simulations on all-to-all networks (N=10) and deterministic scale-free networks (N=81) showing that the average order parameter increases with coupling strength for several values of natural-frequency dispersion (Figures 3 and 6). The conclusion claims that this physics-informed approach establishes a rigorous mathematical foundation for designing, analyzing, and optimizing scalable, adaptive, and interpretable multi-agent AI systems.

Significance. If the proposed mapping were valid, the paper would offer a quantitative, physics-based design tool for agentic AI orchestration, and the explicit model statement plus the code availability in footnote [36] are useful starting points. The internal mathematics is straightforward and appears internally consistent, and the reported monotonic increase of the average order parameter with coupling strength is plausible in the tested parameter ranges. However, the central analogy is uncalibrated and structurally questionable, and the simulations do not distinguish the model's behavior from standard Kuramoto physics or from amplitude-weighting artifacts. The paper therefore does not currently establish its headline claims, although the interdisciplinary analogy may be of some pedagogical interest.

major comments (3)
  1. [Section II.B, Eq. (3)] The paper's central claim rests on identifying the oscillator phase theta_i with task progress and the amplitude r_i with influence or resources (Sections II.B and II.H). This mapping is load-bearing but untested: no calibration against any real multi-agent LLM system or human-agent team is provided. Structurally, theta_i is periodic, so theta_i and theta_i + 2*pi are the same oscillator state, whereas task progress is monotone and bounded, and the sine coupling sin(theta_j - theta_i) depends only on cyclic phase differences, not on absolute progress or on the content of exchanged messages. The Chain-of-Thought correspondence in Section I.A is a list of qualitative parallels rather than a formal derivation. Because the abstract and conclusion claim a 'rigorous mathematical foundation,' this unvalidated analogy is a load-bearing gap that is not addressed by the simulations.
  2. [Section III.A, Figures 3 and 6] The main quantitative finding, that the average order parameter increases with coupling strength, is reported without error bars, without the number of independent runs, and without comparison with analytic Kuramoto results or with a baseline amplitude-free or unweighted model. Since the natural frequencies omega_i are random samples, the average order parameter is a random variable; the curves in Figures 3 and 6 therefore do not support the word 'robustly' as used in the abstract. Without baselines, the observed monotonic increase is exactly the standard Kuramoto behavior and provides no evidence specific to multi-agent AI systems.
  3. [Section II.C, Eq. (4)] The order parameter is amplitude-weighted, R(t) = |(1/N) sum_j r_j e^{i theta_j}|. Because the radial dynamics in Eq. (3) also depend on the phase differences through the cosine coupling, an increase in the average R with epsilon may reflect changes in the amplitudes r_j rather than genuine phase coherence. The paper does not report the unweighted phase coherence or the distribution of r_j over time, so the claimed synchronization enhancement is not separated from the construction of the model. This weakens the interpretation of R(t) as a measure of task completion.
minor comments (5)
  1. [Figure 5 caption] There is a typo: 'determinstic' should be 'deterministic'.
  2. [Section II.C] The statement that R(t) ranges from 0 to 1 holds only under additional assumptions on the amplitudes r_j (for example, r_j <= 1); these assumptions are not stated and can be violated for lambda > 1.
  3. [Figures 3 and 6] The captions should specify how the average order parameter is computed, including the time window used, whether transients were discarded, and the number of independent realizations.
  4. [Footnote [36]] The code availability statement, 'available on GitHub (subject to request and permission)', is ambiguous for reproducibility; a direct repository link and a clear license would be preferable.
  5. [Reference [20]] Reference [20] appears to be an editorial reprint volume rather than a substantive methodological source; the authors should cite the original journal articles that the collection summarizes.

Circularity Check

1 steps flagged · score 3.0 of 10

Self-definitional 'demonstration': increased coupling raises R because ε is defined as the phase-cohering coupling and R as phase coherence; AI-relevance claims remain analogical rather than derived.

  1. self definitional [Section II.B (Eq. 3), Section II.C (Eq. 4), and Section III.A (Fig. 3)]
    "Interestingly, Fig. 3 shows the relationship of the average order parameter (⟨R⟩) with the coupling strength (ϵ), for different values of standard deviation of the natural frequency of agents, σ∈{ 0.1, 0.5, 1.0, 2.0}. It clearly demonstrates that the network synchronizes better with increasing coupling strength despite heterogeneity between individual agents in the system."

    In Eq. (3), ε is defined as the prefactor multiplying the phase-aligning term r_j sin(θ_j−θ_i), so ε is by construction the strength of the synchronizing interaction; in Eq. (4), R is defined as the magnitude of the complex mean of the phases, so R is by construction a phase-coherence measure. The Section III finding that larger ε yields larger ⟨R⟩ is therefore a direct consequence of these definitions, not an independent prediction obtained from multi-agent AI data. Presenting this model-generated curve as evidence that 'the network synchronizes better with increasing coupling strength' reduces the result to the model's construction.

full rationale

The paper is primarily an analogical modeling paper: Eq. (3) is a standard Kuramoto-type phase-amplitude system, and no parameter is fitted to any real multi-agent LLM or agent system. The simulations in Figs. 2, 3, 5, and 6 are internally consistent integrations of the stated equations, so there is no fitted-input-called-prediction circularity and no hidden statistical forcing. The sole self-citation, Ref. [38], is used only to support the choice of a deterministic scale-free test network and is not load-bearing for the central claim. The main circularity concern is the self-definitional status of the central demonstration: because ε enters Eq. (3) as the strength of the phase-cohering sine coupling and R in Eq. (4) is defined as the phase-coherence magnitude, the paper's headline result that increasing coupling raises synchronization is built into the definitions rather than established empirically. The Chain-of-Thought correspondence in Section I.A is an explicit list of qualitative parallels, not a derivation, so it is an unsupported analogy rather than a circular step. Overall, the AI-relevance claims should be read as a proposed interpretation of an established dynamical model, with the empirical validation still missing.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities, but it introduces several free modeling choices (λ, ε, σ, network construction parameters) and postulates an analogy between AI agents and oscillators without external validation. The central result (coupling increases synchronization) is a property of the chosen model, so the ledger is dominated by the analogical mapping and the arbitrary parameter choices rather than by fitted constants.

free parameters (4)
  • λ (amplitude growth parameter) = 1.0 (all runs)
    Sets the equilibrium amplitude of an isolated agent to √λ; chosen by hand as an illustrative value, not derived from any AI agent data.
  • ε (coupling strength) = 5.0 for all-to-all, 30.0 for scale-free; swept in ⟨R⟩ vs ε curves
    The key control parameter for the synchronization claim; values are arbitrary illustrative choices and the sweep range is not justified by application constraints.
  • σ (std of natural frequencies) = 0.05 to 2.0 across runs
    Controls agent heterogeneity; the specific values are a scan range with no grounding in measured AI processing speeds.
  • Deterministic scale-free network parameters (replica number or hierarchy levels) = Not stated (N=81)
    The scale-free topology generation parameters from [37,38] are not reported, so the exact network in Fig. 4 is under-specified.
assumptions (4)
  • domain assumption The Kuramoto model with amplitude dynamics (Eq. 3) is a faithful representation of multi-agent AI collaboration.
    No empirical evidence connects agent progress to oscillator phase or agent influence to r_i; the mapping is posited in Sections II.B and II.H.
  • domain assumption The Chain-of-Thought reasoning process corresponds one-to-one to synchronization dynamics.
    Section I.A lists parallels (iterative steps, context, convergence) as a correspondence, but this is not derived or tested against LLM behavior.
  • ad hoc to paper The intrinsic amplitude term r_i(λ - r_i²) is the correct way to model agent resource or influence dynamics.
    This logistic-growth form is chosen for its saturation property, not derived from AI operational constraints; no justification is given for its functional form.
  • standard math Standard results of Kuramoto synchronization (e.g., monotone increase of order parameter with coupling) are assumed.
    The paper does not derive the synchronization transition analytically; it relies on established Kuramoto theory, which is cited generally in [16].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Synchronization Dynamics of Heterogeneous, Collaborative Multi-Agent AI Systems." pith.science (2026). https://pith.science/paper/3RH4FBXW

@misc{pith2026250812314,
  author       = {Pith},
  title        = {Pith review of: Synchronization Dynamics of Heterogeneous, Collaborative Multi-Agent AI Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3RH4FBXW}},
  note         = {Machine review of arXiv:2508.12314}
}
read the original abstract

We present a novel interdisciplinary framework that bridges synchronization theory and multi-agent AI systems by adapting the Kuramoto model to describe the collective dynamics of heterogeneous AI agents engaged in complex task execution. By representing AI agents as coupled oscillators with both phase and amplitude dynamics, our model captures essential aspects of agent specialization, influence, and communication within networked systems. We introduce an order parameter to quantify the degree of coordination and synchronization, providing insights into how coupling strength, agent diversity, and network topology impact emergent collective behavior. Furthermore, we formalize a detailed correspondence between Chain-of-Thought prompting in AI reasoning and synchronization phenomena, unifying human-like iterative problem solving with emergent group intelligence. Through extensive simulations on all-to-all and deterministic scale-free networks, we demonstrate that increased coupling promotes robust synchronization despite heterogeneous agent capabilities, reflecting realistic collaborative AI scenarios. Our physics-informed approach establishes a rigorous mathematical foundation for designing, analyzing, and optimizing scalable, adaptive, and interpretable multi-agent AI systems. This work opens pathways for principled orchestration of agentic AI and lays the groundwork for future incorporation of learning dynamics and adaptive network architectures to further enhance system resilience and efficiency.

Figures

Figures reproduced from arXiv: 2508.12314 by the authors.

Figure 1
Figure 1. FIG. 1. (Color online) Illustration of a complete network of [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. (Color online) Temporal evolution of (a) phases. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. (Color online) Relationship of the average order pa [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: FIG. 4. (Color online) Network topology of the undirected [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 6
Figure 6. Figure 6: FIG. 6. (Color online) Relationship of the average order pa [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 5
Figure 5. Figure 5: FIG. 5. (Color online) Temporal evolution of (a) phases. (b) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 18 canonical work pages

  1. [36]

    The code and data supporting this study are available on GitHub (subject to request and permission) at: https: //github.com/chiranjitmitra/sync_ai_agents

    work page

  2. [1]

    S. H. Strogatz, Exploring complex networks, Nature 410, 268 (2001)

    2001

  3. [2]

    Albert and A.-L

    R. Albert and A.-L. Barab´ asi, Statistical mechanics of complex networks, Reviews of Modern Physics 74, 47 (2002)

    2002

  4. [3]

    S. N. Dorogovtsev and J. F. Mendes, Evolution of net- works, Advances in Physics 51, 1079 (2002)

    work page 2002

  5. [4]

    M. E. Newman, The Structure and Function of Complex Networks, SIAM Review 45, 167 (2003)

    work page 2003

  6. [5]

    Boccaletti, V

    S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D.- U. Hwang, Complex networks: Structure and dynamics, Physics Reports 424, 175 (2006)

    2006

  7. [6]

    Newman, Networks: An Introduction (Oxford Uni- versity Press, New York, 2010)

    M. Newman, Networks: An Introduction (Oxford Uni- versity Press, New York, 2010)

    work page 2010

  8. [7]

    Casper, L

    S. Casper, L. Bailey, R. Hunter, C. Ezell, E. Ca- bal´ e, M. Gerovitch, S. Slocum, K. Wei, N. Jurkovic, A. Khan, et al. , The AI Agent Index, arXiv preprint arXiv:2502.01635 (2025), arXiv:2502.01635

    arXiv 2025

Show all 38 references
  1. [8]

    Bousetouane, Agentic Systems: A Guide to Trans- forming Industries with Vertical AI Agents, arXiv preprint arXiv:2501.00881 (2025), arXiv:2501.00881

    F. Bousetouane, Agentic Systems: A Guide to Trans- forming Industries with Vertical AI Agents, arXiv preprint arXiv:2501.00881 (2025), arXiv:2501.00881

    2025 arXiv

  2. [9]

    Mukherjee and H

    A. Mukherjee and H. H. Chang, Agentic AI: Auton- omy, Accountability, and the Algorithmic Society, arXiv preprint arXiv:2502.00289 (2025), arXiv:2502.00289

    2025 arXiv

  3. [10]

    Okpala, A

    I. Okpala, A. Golgoon, and A. R. Kannan, Agentic AI Systems Applied to tasks in Financial Services: Model- ing and model risk management crews, arXiv preprint arXiv:2502.05439 (2025), arXiv:2502.05439

    2025 arXiv

  4. [11]

    J. J. Hopfield, Neural networks and physical systems with emergent collective computational abilities, Proceedings of the National Academy of Sciences 79, 2554 (1982)

    1982

  5. [12]

    D. J. Amit, H. Gutfreund, and H. Sompolinsky, Statisti- cal mechanics of neural networks near saturation, Annals of Physics 173, 30 (1987)

    1987

  6. [13]

    Tkacik, E

    G. Tkacik, E. Schneidman, M. J. Berry II, and W. Bialek, Spin glass models for a network of real neurons, arXiv preprint arXiv:0912.5409 (2009), arXiv:0912.5409

    2009 arXiv

  7. [14]

    Raissi, P

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, Physics- informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational Physics 378, 686 (2019)

    2019

  8. [15]

    Pikovsky, M

    A. Pikovsky, M. Rosenblum, and J. Kurths, Synchroniza- tion: A Universal Concept in Nonlinear Sciences , Vol. 12 (Cambridge University Press, Cambridge, 2003)

    2003

  9. [16]

    F. A. Rodrigues, T. K. D. Peron, P. Ji, and J. Kurths, The Kuramoto model in complex networks, Physics Re- ports 610, 1 (2016)

    2016

  10. [17]

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V. Le, D. Zhou, et al. , Chain-of-Thought Prompting Elicits Reasoning in Large Language Mod- els, Advances in Neural Information Processing Systems 35, 24824 (2022)

    2022

  11. [18]

    Kojima, S

    T. Kojima, S. S. Gu, M. Reid, Y. Matsuo, and Y. Iwa- sawa, Large Language Models are Zero-Shot Reasoners, Advances in Neural Information Processing Systems 35, 22199 (2022)

    2022

  12. [19]

    K.-T. Tran, D. Dao, M.-D. Nguyen, Q.-V. Pham, B. O’Sullivan, and H. D. Nguyen, Multi-Agent Collab- oration Mechanisms: A Survey of LLMs, arXiv preprint arXiv:2501.06322 (2025), arXiv:2501.06322

    2025 arXiv

  13. [20]

    Hu and Z

    J. Hu and Z. Peng, Mathematical Methods Applied in Artificial Intelligence and Multi-Agent Systems (MDPI- Multidisciplinary Digital Publishing Institute, Basel, 9 2024)

    2024

  14. [21]

    Y. Xiao, G. Shi, and P. Zhang, Towards Agentic AI Networking in 6G: A Generative Foundation Model- as-Agent Approach, arXiv preprint arXiv:2503.15764 (2025), arXiv:2503.15764

    2025 arXiv

  15. [22]

    Agranat and M

    R. Agranat and M. S. Gal, Fueling Concentration: Net- work Effects and AI Agents, Network Law Review, Spring , 2016 (2025)

    2025

  16. [23]

    Q. Wu, G. Bansal, J. Zhang, Y. Wu, B. Li, E. Zhu, L. Jiang, X. Zhang, S. Zhang, J. Liu, et al. , AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation, in First Conference on Language Model- ing (2024)

    2024

  17. [24]

    Ni and M

    B. Ni and M. J. Buehler, MechAgents: Large language model multi-agent collaborations can solve mechanics problems, generate new data, and integrate knowledge, Extreme Mechanics Letters 67, 102131 (2024)

    2024

  18. [25]

    C. S. de Witt, Open Challenges in Multi-Agent Security: Towards Secure Systems of Interacting AI Agents, arXiv preprint arXiv:2505.02077 (2025), arXiv:2505.02077

    2025 arXiv

  19. [26]

    Sen and S

    P. Sen and S. M. Jakkaraju, Modeling AI-Human Col- laboration as a Multi-Agent Adaptation, arXiv preprint arXiv:2504.20903 (2025), arXiv:2504.20903

    2025 arXiv

  20. [27]

    H. Du, S. Thudumu, R. Vasa, and K. Mouzakis, A Survey on Context-Aware Multi-Agent Systems: Tech- niques, Challenges and Future Directions, arXiv preprint arXiv:2402.01968 (2024), arXiv:2402.01968

    2024 arXiv

  21. [28]

    Z. Xi, W. Chen, X. Guo, W. He, Y. Ding, B. Hong, M. Zhang, J. Wang, S. Jin, E. Zhou, et al., The rise and potential of large language model based agents: a survey, Science China Information Sciences 68, 121101 (2025)

    2025

  22. [29]

    T. Guo, X. Chen, Y. Wang, R. Chang, S. Pei, N. V. Chawla, O. Wiest, and X. Zhang, Large Language Model based Multi-Agents: A Survey of Progress and Challenges, arXiv preprint arXiv:2402.01680 (2024), arXiv:2402.01680

    2024 arXiv

  23. [30]

    S. Han, Q. Zhang, Y. Yao, W. Jin, and Z. Xu, LLM Multi-Agent Systems: Challenges and Open Problems, arXiv preprint arXiv:2402.03578 (2024), arXiv:2402.03578

    2024 arXiv

  24. [31]

    A. Chan, C. Ezell, M. Kaufmann, K. Wei, L. Hammond, H. Bradley, E. Bluemke, N. Rajkumar, D. Krueger, N. Kolt, et al., Visibility into AI Agents, in Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (2024) pp. 958–973

    2024

  25. [32]

    Durante, Q

    Z. Durante, Q. Huang, N. Wake, R. Gong, J. S. Park, B. Sarkar, R. Taori, Y. Noda, D. Terzopoulos, Y. Choi, et al. , Agent AI: Surveying the Horizons of Multimodal Interaction, arXiv preprint arXiv:2401.03568 (2024), arXiv:2401.03568

    2024 arXiv

  26. [33]

    Lanham, AI Agents in Action (Manning Publications, New York, 2025)

    M. Lanham, AI Agents in Action (Manning Publications, New York, 2025)

    2025

  27. [34]

    Sapkota, K

    R. Sapkota, K. I. Roumeliotis, and M. Karkee, AI Agents vs. Agentic AI: A Conceptual Taxonomy, Applications and Challenges, arXiv preprint arXiv:2505.10468 (2025), arXiv:2505.10468

    2025

  28. [35]

    S. Kim, Y. Yu, and H. Seo, Artificial intelligence or- chestration for text-based ultrasonic simulation via self- review by multi-large language model agents, Scientific Reports 15, 12474 (2025)

    2025

  29. [37]

    Barab´ asi, E

    A.-L. Barab´ asi, E. Ravasz, and T. Vicsek, Deterministic scale-free networks, Physica A: Statistical Mechanics and its Applications 299, 559 (2001)

    2001

  30. [38]

    Mitra, A

    C. Mitra, A. Choudhary, S. Sinha, J. Kurths, and R. V. Donner, Multiple-node basin stability in complex dynam- ical networks, Physical Review E 95, 032317 (2017)

    2017

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.