Pith. sign in

REVIEW 4 major objections 6 minor 58 references

Why do AI agents communicate in human language?

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Natural language is the wrong medium for AI-to-AI coordination, this paper argues.

desk verdict A useful research agenda that overstates its formal case: the paper's central claim of inevitable cascading semantic loss is asserted, not proven. read the letter →

arxiv 2506.02739 v1 pith:Z2WSMZUQ submitted 2025-06-03 cs.AI

classification cs.AI
keywords multi-agentsystemslargelanguagemodelsnaturalcommunicationsemanticmisalignmentcascadinglossrolepersistencestructurednext-tokenprediction
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

AI agents that coordinate by sending natural-language messages are, on this paper's view, pushing their high-dimensional internal states through a narrow, lossy pipe. The authors argue that the semantic space of natural language is structurally misaligned with the vector spaces in which LLMs actually reason, so every exchange compresses away information that cannot be recovered, and the errors accumulate across turns into behavioral drift, role confusion, and broken task chains. Because current LLMs are trained only to predict the next token over linear text streams, the paper concludes that they cannot support scalable multi-agent coordination without an architectural change. The proposed remedy is a native multi-agent modeling paradigm in which communication is structured tensor-state exchange, role identities are bound to internal states, and perception, planning, and coordination are decoupled.

What carries the argument

The load-bearing object is the generation map $f: \mathcal{H} \to \mathcal{L}$, where $\mathcal{H}$ is the high-dimensional continuous semantic state space of an LLM and $\mathcal{L}$ is the discrete space of natural-language token sequences. The paper treats $f$ as a many-to-one, non-invertible compression, so that message transmission is modeled as $h^{(j)}_t \mapsto f(h^{(j)}_t) \mapsto \tilde{f}^{-1}(f(h^{(j)}_t)) \neq h^{(j)}_t$, and it formalizes the accumulated error as $L_{\text{cascade}} = \sum_{t=1}^{T} \mathbb{E}_{j \in A}[\|h^{(j)}_t - \hat{h}^{(j)}_t\|^2]$. Everything else—the need for role-space binding $h^{(i)}_t \sim p(h \mid r^{(i)})$, structured message tensors $m^{(t)}_{ji} = G(h^{(j)}_t)$, and the coordination graph $\mathcal{G}_t = (\mathcal{A}, \mathcal{R}, \mathcal{E}_t)$—is presented as the remedy once this loss is acknowledged to have a nonzero lower bound.

What would settle it

Sample many pairs of distinct internal states that map to the same utterance, feed the utterance to a second model, and check whether the recovered state is closer to the true sender state than chance; if the approximate inverse $\tilde{f}^{-1}$ recovers the sender state accurately, the many-to-one and non-invertibility premise fails. A second check: measure whether task success in a long-horizon two-agent protocol drops monotonically with conversation length when context windows and memory are held fixed; if no such drop appears, the claimed nonzero lower bound on $L_{\text{cascade}}$ is not supported.

Watch

Extended reading notes

Core claim

The central claim is that natural language is not merely an inefficient protocol for agent-to-agent communication but structurally the wrong medium. The paper models each agent's internal semantic state as a point in a high-dimensional tensor space $\mathcal{H}$ and the language it emits as a projection $f: \mathcal{H} \to \mathcal{L}$ into a discrete token space. Because $f$ is claimed to be many-to-one and non-invertible, distinct internal states collapse to identical utterances, and a receiving agent's reconstruction $\hat{h} = \tilde{f}^{-1}(l)$ almost never equals the sender's state $h$; this per-round error accumulates as a cascading semantic loss with a nonzero lower bound. The same structural mismatch, the paper argues, explains why LLM agents show goal drift, pseudo-execution, role confusion, and broken task chains, and it cannot be fixed by prompting, memory, or better protocols. The positive thesis is that these failures are architectural and must be addressed by training a new class of native multi-agent models with role persistence, structured tensor communication, inter-agent state synchronization, and decoupled perception, planning, and coordination modules.

Load-bearing premise

Everything rests on the premise that an agent's internal state cannot be recovered from the language it emits; if models can learn near-invertible mappings from text back to internal state, the claimed cascade of semantic loss would not be inevitable.

Editorial extensions

If this is right

  • Long-horizon, multi-role agent teams built on current LLMs will accumulate state divergence, so extended task chains should be expected to degrade even when each individual reply looks coherent.
  • Prompt tuning, memory buffers, and protocol wrappers can reduce symptoms but cannot remove the underlying loss, because the bottleneck is the generation architecture itself.
  • A workable fix requires models whose role identifiers are bound to internal states, whose communication runs on structured tensors rather than tokens, and whose action policies are decoupled from language output.
  • Hybrid interfaces will be needed so that natural language remains a human-debuggable channel while the coordination path operates on state-aligned structured messages.

Reading between the lines

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

  • The paper's compression logic implies the same loss should be measurable in a single agent: a model that states a plan in words and later executes it should show measurable divergence between the original plan state and the state recovered from its own paraphrase.
  • An empirical test the paper does not run would compare task-completion rates over matched interaction budgets between language-only agent teams and teams sharing latent-state communication channels; a large gap would support the proposal, while no gap would undermine it.
  • The argument sets a correctness ceiling below communication bandwidth: even arbitrarily large models would keep losing state through the discrete-token bottleneck unless training objectives change, not just parameters.
  • The proposed paradigm suggests new benchmarks built around multi-turn coordination tasks that vary conversation length and role count and measure state-recovery error rather than only final task success.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. This position paper argues that natural language is structurally misaligned with the high-dimensional vector spaces in which LLMs operate, so using natural language for inter-agent communication causes irreversible information loss, semantic drift, and cumulative coordination errors. The authors formalize the generation process as a projection f: H -> L from an internal semantic space to a discrete token space, claim that f is many-to-one and non-invertible, and use this to assert that the cascading semantic loss L_cascade has a non-zero lower bound. They further argue that the next-token prediction training objective lacks structural support for role persistence, task decomposition, and inter-agent state synchronization, and they propose a native multi-agent modeling paradigm with role persistence, structured tensor communication, state synchronization, and functional decoupling. The paper is primarily an argumentative survey: it contains no new experiments, no machine-checked proofs, and its central formal claims rest on unproven assumptions.

Significance. The question the paper raises is timely and practically important: if natural-language communication between LLM-based agents is fundamentally lossy, then scalable multi-agent systems may require new communication protocols or even new training paradigms. The paper offers a useful synthesis of known failure modes, a clear set of design requirements in Section IV, and a helpful discussion of open challenges in Section IV.C. At the same time, the paper's central formal claim is not established: the alleged inevitability of cascading semantic loss is built into the definition of f rather than derived, and the architectural incompatibility claim is asserted rather than demonstrated. The paper is therefore best viewed as a research agenda or position statement rather than a completed technical result, and its current form overstates the conclusiveness of its own analysis.

major comments (4)
  1. [Section III.A, Eqs. (1)-(3)] The claimed non-zero lower bound on L_cascade is not derived; it is assumed. Eq. (1) defines f as many-to-one, and Eq. (3) defines L_cascade as the sum of per-step state-recovery errors. The text then asserts that because f is many-to-one, the error delta_t accumulates and L_cascade has a non-zero lower bound. This does not follow: any finite-rate channel from a continuous space is many-to-one, and the relevant question is whether the lost information is task-relevant and whether an approximate inverse g: L -> H can recover the task-relevant subspace. No argument is given that the equivalence classes of f cross task-relevant distinctions, nor that g cannot be learned through shared training or fine-tuning. Without such an argument, the conclusion that cascading semantic loss is inevitable is unsupported.
  2. [Section III.A, Eq. (2)] The formalization conflates sender and receiver states and assumes the receiver's update is T(f^{-1}(f(h_t))). The error term delta_t is defined as the difference between h_t and its reconstruction from the message, but the claim that this error 'accumulates over time' is an assertion, not a consequence of the equation. If the receiver can maintain a conversation-level state, or if the task depends only on a low-dimensional projection of h, the reconstruction error may be irrelevant or non-accumulating. The paper needs either an explicit counterexample with concrete dynamics or a formal condition under which accumulation is guaranteed; otherwise Eq. (3) merely restates the assumption built into Eq. (1).
  3. [Section III.C, Eq. (6)] The claim that the next-token prediction objective 'lacks structural priors' and therefore cannot support role persistence, task boundaries, or multi-agent dependencies is a non-sequitur. A model trained by next-token prediction can in principle learn to maintain role-consistent states from conversational data; the absence of an explicit inductive bias does not imply that the relevant structure cannot be learned. The paper provides no empirical or theoretical evidence that such structures are unlearnable under the standard objective. This is load-bearing because the architectural incompatibility argument motivates the proposed shift to a new training paradigm; as written, it remains an unsupported assertion.
  4. [Section III.B and III.C, Eqs. (4)-(5)] The 'Protocol-Induced Agent Misbehavior' and 'Architectural Incompatibility' arguments repeatedly rely on the same unproved assertion that f is lossy and non-invertible. The empirical examples (AutoGPT, AgentVerse, Cemri et al.) are suggestive, but they are not controlled comparisons between natural language and an alternative protocol; they are consistent with the paper's thesis but do not establish that the cause is structural misalignment rather than limited context windows, insufficient prompting, or other engineering factors. To make this causal claim load-bearing, the paper would need a controlled study or at least a formal model with clearly stated assumptions and testable predictions.
minor comments (6)
  1. [Section II.B and Ref. [9]] The paper refers to 'MCP' as 'Multi-Agent Communication Protocol' in the introduction, but Ref. [9] is the Model Context Protocol; the acronym is used inconsistently and should be corrected.
  2. [Section II.C] The text attributes a claim about lack of explicit state modeling to 'Microsoft Research [14]', but Ref. [14] is by Laban et al.; the attribution is inaccurate.
  3. [Section III.A, Eq. (2)] The subscripts in Eq. (2) are inconsistent: the left side is h(i)_{t+1} while the right side uses h(j)_t, and the roles of i and j are not defined. Please clarify the state-update notation.
  4. [Section III.A] The statement that L_cascade 'admits a non-zero lower bound and cannot be eliminated under standard decoding assumptions' is not accompanied by a precise statement of what those 'standard decoding assumptions' are; the assumptions should be stated explicitly or the 'theoretically' claim should be removed.
  5. [Section III.C] The direct-sum decomposition S = \bigoplus S^{(i)} with T(S^{(i)}) subset of S^{(i)} is introduced as an 'ideal' condition, but no justification is given for why such a decomposition is necessary or achievable, and the direct-sum notation is never defined.
  6. [References] Several claims are supported by citations to blog posts and GitHub issues (e.g., Refs. [16], [55], [56]); these are less authoritative than peer-reviewed sources and should be supplemented or clearly flagged as anecdotal.

Circularity Check

1 steps flagged · score 6.0 of 10

The formal lower bound on cascading semantic loss restates the assumed many-to-one-ness of Eq. (1); the inevitability claim is definitional, while external empirical work supports the system-level symptoms.

  1. self definitional [Section III.A, Eq. (1) and Eq. (3)]
    "As f is typically many-to-one, distinct internal semantic states may correspond to identical linguistic expressions, leading to semantic aliasing and non-invertibility. ... Given the inevitability of semantic aliasing and linguistic ambiguity in current LLM-based communication systems, this loss admits a non-zero lower bound and cannot be eliminated under standard decoding assumptions."

    Eq. (3)'s L_cascade is defined as a sum of per-step reconstruction errors. The claimed nonzero lower bound is then justified by 'the inevitability of semantic aliasing,' which is exactly the Eq. (1) assertion that f is many-to-one and non-invertible. There is no independent derivation: if f is non-invertible and reception is written as f^{-1}, then some h must satisfy \hat h = f^{-1}(f(h)) != h, making the summed error positive by construction. The paper does not prove the aliased directions contain task-relevant state, nor that a shared-trained inverse g:L→H could not keep task error small; finite-rate lossiness alone does not imply semantic drift. So the 'theoretical' inevitability is a restatement of the assumption, not a result derived from it.

full rationale

The only formal 'result' is the nonzero lower bound on L_cascade in Eq. (3), and it is self-definitional: it follows directly from declaring f in Eq. (1) many-to-one/non-invertible and writing decoding as f^{-1}. The paper does not quantify the bound, prove positivity under a task-relevant metric, or exclude learned approximate inverses; a finite-rate channel from a continuous space is always many-to-one, so such lossiness alone cannot establish behavioral drift. This is the one circular load-bearing step. The rest of the paper is a synthesis of empirical failures (lost-in-conversation, pseudo-execution, role confusion) from external sources, which provides independent support for the symptoms though not for the claimed theoretical inevitability. The only self-citation [12] appears in the introduction for human-in-the-loop supervision and is not load-bearing. No fitted-parameter-as-prediction or author-imported uniqueness theorem appears. Because the central formal claim reduces to its premise, the score is 6 rather than 0-2; because the practical concerns are externally documented, it is not an 8-10.

Assumptions & free parameters 1 free parameters · 4 assumptions · 3 invented entities

All substantive claims about misalignment and architectural incompatibility rely on unverified assumptions about the nature of LLM internal spaces, the non-invertibility of language generation, and the requirements of coordination. The paper proposes novel constructs (role spaces, message tensors, coordination graphs) without providing independent evidence or falsifiable predictions.

free parameters (1)
  • epsilon (alignment error bound) = unspecified
    Introduced in Eq. (9) as a design target for inter-agent synchronization; no procedure for setting its value is given, so it acts as a free tolerance.
assumptions (4)
  • domain assumption Natural language is a discrete symbolic system with sparse and non-differentiable transitions, while LLM internal representations are continuous, dense tensors.
    Section III.A asserts this dichotomy without empirical support; the misalignment argument rests on it.
  • ad hoc to paper The language generation map f: H -> L is many-to-one and non-invertible.
    Defined in Eq. (1); the conclusion that communication via language incurs irreducible loss is a direct consequence of this definition, not an independent finding.
  • domain assumption Communication among agents requires structural isomorphism between the communication medium and internal representations.
    Stated in Section III.A as a presupposition; if coordination can tolerate lossy communication, the central problem weakens.
  • domain assumption The next-token prediction objective cannot support role persistence, task boundaries, or inter-agent dependencies.
    Section III.C argues this without a formal proof or comparative experiments; it is a plausible but unverified architectural claim.
invented entities (3)
  • Role space R = {r^(1), ..., r^(n)} with h^(i)_t ~ p(h|r^(i))
    purpose: Bind each agent's internal state to a stable role identity across turns.
    Proposed as a design requirement in Section IV.A; no implementation or test is provided.
  • Structured message tensor M_ij = G(h_j)
    purpose: Replace natural-language messages with executable state exchanges.
    Defined in Eq. (8) as a proposal; no evidence that such tensors can be learned or that they outperform language.
  • Coordination graph G_t = (A, R, E_t)
    purpose: Explicitly track role relations, task dependencies, and semantic boundaries among agents.
    Proposed in Section IV.B, Eq. (10); no benchmark or falsifiable prediction is supplied.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Why do AI agents communicate in human language?." pith.science (2026). https://pith.science/paper/Z2WSMZUQ

@misc{pith2026250602739,
  author       = {Pith},
  title        = {Pith review of: Why do AI agents communicate in human language?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z2WSMZUQ}},
  note         = {Machine review of arXiv:2506.02739}
}
read the original abstract

Large Language Models (LLMs) have become foundational to modern AI agent systems, enabling autonomous agents to reason and plan. In most existing systems, inter-agent communication relies primarily on natural language. While this design supports interpretability and human oversight, we argue that it introduces fundamental limitations in agent-to-agent coordination. The semantic space of natural language is structurally misaligned with the high-dimensional vector spaces in which LLMs operate, resulting in information loss and behavioral drift. Beyond surface-level inefficiencies, we highlight a deeper architectural limitation: current LLMs were not trained with the objective of supporting agentic behavior. As such, they lack mechanisms for modeling role continuity, task boundaries, and multi-agent dependencies. The standard next-token prediction paradigm fails to support the structural alignment required for robust, scalable agent coordination. Based on this, we argue that two core questions deserve careful examination: first, given that AI agents fundamentally operate in high-dimensional vector spaces, should they rely on a language system originally designed for human cognition as their communication medium? Second, should we consider developing a new model construction paradigm that builds models from the ground up to natively support structured communication, shared intentionality, and task alignment in multi-role, multi-agent environments? This paper calls for a reconsideration not only of how agents should communicate, but also of what it fundamentally means to train a model that natively supports multi-agent coordination and communication.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 31 canonical work pages

  1. [1]

    Multi-agent collaboration in ai: Enhancing software development with autonomous llms,

    M. Wasif and D. Tunkel, “Multi-agent collaboration in ai: Enhancing software development with autonomous llms,” 2025

  2. [2]

    Ai agents: from co-pilot to autopilot,

    Financial Times, “Ai agents: from co-pilot to autopilot,” https://www.ft. com/content/3e862e23-6e2c-4670-a68c-e204379fe01f, 2025, accessed: 2025-05-28

  3. [3]

    Large language models for multi- robot systems: A survey,

    P. Li, Z. An, S. Abrar, and L. Zhou, “Large language models for multi- robot systems: A survey,”arXiv preprint arXiv:2502.03814, 2025

  4. [4]

    What is a multi-agent system,

    Relevance AI, “What is a multi-agent system,” 2024, accessed: 2025-05-28. [Online]. Available: https://relevanceai.com/ learn/what-is-a-multi-agent-system

  5. [5]

    Multi-agent system — wikipedia, the free encyclopedia,

    Wikipedia contributors, “Multi-agent system — wikipedia, the free encyclopedia,” https://en.wikipedia.org/wiki/Multi-agent_system, 2025, accessed: 2025-05-28

  6. [6]

    Llms for multi-agent cooperation,

    X. Lyu, “Llms for multi-agent cooperation,” https://xue-guang.com/post/ llm-marl/, 2025, accessed: 2025-05-28

  7. [7]

    A survey of agent interoperability protocols: Model context protocol (mcp), agent communication protocol (acp), agent-to-agent protocol (a2a), and agent network protocol (anp),

    A. Ehtesham, A. Singh, G. K. Gupta, and S. Kumar, “A survey of agent interoperability protocols: Model context protocol (mcp), agent communication protocol (acp), agent-to-agent protocol (a2a), and agent network protocol (anp),”arXiv preprint arXiv:2505.02279, 2025

  8. [8]

    Announcing the agent2agent proto- col (a2a),

    Google, “Announcing the agent2agent proto- col (a2a),” https://developers.googleblog.com/en/ a2a-a-new-era-of-agent-interoperability/, 2025, accessed: 2025-05-29

Show all 58 references
  1. [9]

    Introducing the model context protocol,

    Anthropic, “Introducing the model context protocol,” https://www. anthropic.com/news/model-context-protocol, 2024, accessed: 2025-05- 29

  2. [10]

    A survey on recent advances in llm-based multi-turn dialogue systems,

    Z. Yi, J. Ouyang, Y . Liu, T. Liao, Z. Xu, and Y . Shen, “A survey on recent advances in llm-based multi-turn dialogue systems,”arXiv preprint arXiv:2402.18013, 2024

  3. [11]

    Multi-agent conversation framework: Autogen 0.2 documentation,

    Microsoft, “Multi-agent conversation framework: Autogen 0.2 documentation,” https://microsoft.github.io/autogen/0.2/docs/Use-Cases/ agent_chat/, 2025, accessed: 2025-05-28

  4. [12]

    Continuum-interaction-driven intelligence: Human-aligned neural architecture via crystallized reasoning and fluid generation,

    P. Zhou, Z. Nie, and H. Li, “Continuum-interaction-driven intelligence: Human-aligned neural architecture via crystallized reasoning and fluid generation,”arXiv preprint arXiv:2504.09301, 2025

  5. [13]

    Evaluating llm-based agents for multi-turn conversations: A survey,

    S. Guan, H. Xiong, J. Wang, J. Bian, B. Zhu, and J.-g. Lou, “Evaluating llm-based agents for multi-turn conversations: A survey,”arXiv preprint arXiv:2503.22458, 2025

  6. [14]

    Llms get lost in multi- turn conversation,

    P. Laban, H. Hayashi, Y . Zhou, and J. Neville, “Llms get lost in multi- turn conversation,”arXiv preprint arXiv:2505.06120, 2025

  7. [15]

    Llm multi-agent systems: Challenges and open problems,

    S. Han, Q. Zhang, Y . Yao, W. Jin, Z. Xu, and C. He, “Llm multi-agent systems: Challenges and open problems,”arXiv preprint arXiv:2402.03578, 2024

  8. [16]

    Why autogpt fails and how to fix it,

    T. Pungas, “Why autogpt fails and how to fix it,” https://www.taivo.ai/ __why-autogpt-fails-and-how-to-fix-it/, 2023, accessed: 2025-05-28

  9. [17]

    Agentverse vs. autogpt: Comparing ai agent development platforms,

    SmythOS, “Agentverse vs. autogpt: Comparing ai agent development platforms,” https://smythos.com/ai-agents/comparison/ agentverse-vs-autogpt/, 2024, accessed: 2025-05-28

  10. [18]

    Why do multi-agent llm systems fail?

    M. Cemri, M. Z. Pan, S. Yang, L. A. Agrawal, B. Chopra, R. Tiwari, K. Keutzer, A. Parameswaran, D. Klein, K. Ramchandranet al., “Why do multi-agent llm systems fail?”arXiv preprint arXiv:2503.13657, 2025

  11. [19]

    Which agent causes task failures and when? on automated failure attribution of llm multi-agent systems,

    S. Zhang, M. Yin, J. Zhang, J. Liu, Z. Han, J. Zhang, B. Li, C. Wang, H. Wang, Y . Chenet al., “Which agent causes task failures and when? on automated failure attribution of llm multi-agent systems,”arXiv preprint arXiv:2505.00212, 2025

  12. [20]

    Language models don’t always say what they think: Unfaithful explanations in chain- of-thought prompting,

    M. Turpin, J. Michael, E. Perez, and S. Bowman, “Language models don’t always say what they think: Unfaithful explanations in chain- of-thought prompting,”Advances in Neural Information Processing Systems, vol. 36, pp. 74 952–74 965, 2023

  13. [21]

    Reasoning models don’t always say what they think,

    Y . Chen, J. Benton, A. Radhakrishnan, J. Uesato, C. Denison, J. Schul- man, A. Somani, P. Hase, M. Wagner, F. Rogeret al., “Reasoning models don’t always say what they think,”arXiv preprint arXiv:2505.05410, 2025

  14. [22]

    Fake alignment: Are llms really aligned well?

    Y . Wang, Y . Teng, K. Huang, C. Lyu, S. Zhang, W. Zhang, X. Ma, Y .-G. Jiang, Y . Qiao, and Y . Wang, “Fake alignment: Are llms really aligned well?”arXiv preprint arXiv:2311.05915, 2023

  15. [23]

    Cut the crap: An economical communication pipeline for llm-based multi-agent systems,

    G. Zhang, Y . Yue, Z. Li, S. Yun, G. Wan, K. Wang, D. Cheng, J. X. Yu, and T. Chen, “Cut the crap: An economical communication pipeline for llm-based multi-agent systems,”arXiv preprint arXiv:2410.02506, 2024

  16. [24]

    Multi-agent ai: Performance metrics & evaluation frameworks,

    G. AI, “Multi-agent ai: Performance metrics & evaluation frameworks,” https://galileo.ai/blog/success-multi-agent-ai, 2025, accessed: 2025-05- 28

  17. [25]

    Emergence of grounded compositional language in multi-agent populations,

    I. Mordatch and P. Abbeel, “Emergence of grounded compositional language in multi-agent populations,” inProceedings of the AAAI conference on artificial intelligence, vol. 32, no. 1, 2018

  18. [26]

    Information theory as a bridge between language function and language form,

    R. Futrell and M. Hahn, “Information theory as a bridge between language function and language form,”Frontiers in Communication, vol. 7, p. 657725, 2022

  19. [27]

    Multi-agent communica- tion meets natural language: Synergies between functional and structural language learning,

    A. Lazaridou, A. Potapenko, and O. Tieleman, “Multi-agent communica- tion meets natural language: Synergies between functional and structural language learning,” inProceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020, pp. 7663–7674

  20. [28]

    Position: Towards a responsible llm-empowered multi-agent systems,

    J. Hu, Y . Dong, S. Ao, Z. Li, B. Wang, L. Singh, G. Cheng, S. D. Ram- churn, and X. Huang, “Position: Towards a responsible llm-empowered multi-agent systems,”arXiv preprint arXiv:2502.01714, 2025

  21. [29]

    Beyond self-talk: A communication-centric survey of llm-based multi- agent systems,

    B. Yan, X. Zhang, L. Zhang, L. Zhang, Z. Zhou, D. Miao, and C. Li, “Beyond self-talk: A communication-centric survey of llm-based multi- agent systems,”arXiv preprint arXiv:2502.14321, 2025

  22. [30]

    Camel: Communicative agents for

    G. Li, H. Hammoud, H. Itani, D. Khizbullin, and B. Ghanem, “Camel: Communicative agents for" mind" exploration of large language model society,”Advances in Neural Information Processing Systems, vol. 36, pp. 51 991–52 008, 2023

  23. [31]

    Crewai: A multi-agent orchestration frame- work for llm agents,

    J. Moura and contributors, “Crewai: A multi-agent orchestration frame- work for llm agents,” https://github.com/joaomdmoura/crewAI, 2024, gitHub Repository

  24. [32]

    Tradingagents: Multi-agents llm financial trading framework,

    Y . Xiao, E. Sun, D. Luo, and W. Wang, “Tradingagents: Multi-agents llm financial trading framework,”arXiv preprint arXiv:2412.20138, 2024

  25. [33]

    An ai- assisted multi-agent dual dialogue system to support mental health care providers,

    O. P. Kampman, Y . S. Phang, S. Han, M. Xing, X. Hong, H. Hoo- sainsah, C. Tan, G. I. Winata, S. Wang, C. Heaukulaniet al., “An ai- assisted multi-agent dual dialogue system to support mental health care providers,”arXiv preprint arXiv:2411.18429, 2024

  26. [34]

    Simulating classroom education with llm-empowered agents,

    Z. Zhang, D. Zhang-Li, J. Yu, L. Gong, J. Zhou, Z. Hao, J. Jiang, J. Cao, H. Liu, Z. Liuet al., “Simulating classroom education with llm-empowered agents,”arXiv preprint arXiv:2406.19226, 2024

  27. [35]

    Chain-of-thought prompting elicits reasoning in large language models,

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhouet al., “Chain-of-thought prompting elicits reasoning in large language models,”Advances in neural information processing systems, vol. 35, pp. 24 824–24 837, 2022

  28. [36]

    Augmented language models: a survey,

    G. Mialon, R. Dessì, M. Lomeli, C. Nalmpantis, R. Pasunuru, R. Raileanu, B. Rozière, T. Schick, J. Dwivedi-Yu, A. Celikyil- mazet al., “Augmented language models: a survey,”arXiv preprint arXiv:2302.07842, 2023

  29. [37]

    Reflex- ion: Language agents with verbal reinforcement learning,

    N. Shinn, F. Cassano, A. Gopinath, K. Narasimhan, and S. Yao, “Reflex- ion: Language agents with verbal reinforcement learning,”Advances in Neural Information Processing Systems, vol. 36, pp. 8634–8652, 2023

  30. [38]

    Kqml as an agent communication language,

    T. Finin, R. Fritzson, D. McKay, and R. McEntire, “Kqml as an agent communication language,” inProceedings of the third international conference on Information and knowledge management, 1994, pp. 456– 463

  31. [39]

    Fipa communicative act library specification,

    T. FIPA, “Fipa communicative act library specification,”Change, vol. 2000, no. 01/18, 2000

  32. [40]

    Wooldridge,An introduction to multiagent systems

    M. Wooldridge,An introduction to multiagent systems. John wiley & sons, 2009

  33. [41]

    Learning to communicate with deep multi-agent reinforcement learning,

    J. Foerster, I. A. Assael, N. De Freitas, and S. Whiteson, “Learning to communicate with deep multi-agent reinforcement learning,”Advances in neural information processing systems, vol. 29, 2016

  34. [42]

    Emergent multi-agent communication in the deep learning era,

    A. Lazaridou and M. Baroni, “Emergent multi-agent communication in the deep learning era,”arXiv preprint arXiv:2006.02419, 2020

  35. [43]

    On the pitfalls of measuring emergent communication,

    R. Lowe, J. Foerster, Y .-L. Boureau, J. Pineau, and Y . Dauphin, “On the pitfalls of measuring emergent communication,” inProceedings of the 18th International Conference on Autonomous Agents and MultiAgent Systems, 2019, pp. 693–701

  36. [44]

    Natural language does not emerge ‘naturally’in multi-agent dialog,

    S. Kottur, J. Moura, S. Lee, and D. Batra, “Natural language does not emerge ‘naturally’in multi-agent dialog,” inProceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, 2017, pp. 2962–2967

  37. [45]

    Multitasking inhibits semantic drift,

    A. P. Jacob, M. Lewis, and J. Andreas, “Multitasking inhibits semantic drift,” inProceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 2021, pp. 5351–5366

  38. [46]

    Comparing agent communication languages and protocols: Choosing the right framework for multi-agent systems,

    SmythOS, “Comparing agent communication languages and protocols: Choosing the right framework for multi-agent systems,” https://smythos.com/developers/ai-agent-development/ agent-communication-languages-and-protocols-comparison/, 2025

  39. [47]

    Cooperate or collapse: Emergence of sustainable coop- eration in a society of llm agents,

    G. Piatti, Z. Jin, M. Kleiman-Weiner, B. Schölkopf, M. Sachan, and R. Mihalcea, “Cooperate or collapse: Emergence of sustainable coop- eration in a society of llm agents,”Advances in Neural Information Processing Systems, vol. 37, pp. 111 715–111 759, 2024

  40. [48]

    A survey on large language model based human-agent systems,

    H. P. Zou, W.-C. Huang, Y . Wu, Y . Chen, C. Miao, H. Nguyen, Y . Zhou, W. Zhang, L. Fang, L. Heet al., “A survey on large language model based human-agent systems,”arXiv preprint arXiv:2505.00753, 2025

  41. [49]

    Are multiagent systems resilient to communication failures?

    P. N. Brown, H. P. Borowski, and J. R. Marden, “Are multiagent systems resilient to communication failures?”arXiv preprint arXiv:1710.08500, 2017

  42. [50]

    Idiosyncrasies in large language models,

    M. Sun, Y . Yin, Z. Xu, J. Z. Kolter, and Z. Liu, “Idiosyncrasies in large language models,”arXiv preprint arXiv:2502.12150, 2025

  43. [51]

    Autogpt+ p: Affordance- based task planning with large language models,

    T. Birr, C. Pohl, A. Younes, and T. Asfour, “Autogpt+ p: Affordance- based task planning with large language models,”arXiv preprint arXiv:2402.10778, 2024

  44. [52]

    The symbol grounding problem,

    S. Harnad, “The symbol grounding problem,”Physica D: Nonlinear Phenomena, vol. 42, no. 1-3, pp. 335–346, 1990

  45. [53]

    Linguistic generalization and compositionality in modern artificial neural networks,

    M. Baroni, “Linguistic generalization and compositionality in modern artificial neural networks,”Philosophical Transactions of the Royal Society B, vol. 375, no. 1791, p. 20190307, 2020

  46. [54]

    How language model hallucinations can snowball,

    M. Zhang, O. Press, W. Merrill, A. Liu, and N. A. Smith, “How language model hallucinations can snowball,” inInternational Conference on Machine Learning. PMLR, 2024, pp. 59 670–59 684

  47. [55]

    Agentic ai: A complete guide to autonomous ai agents,

    VideoSDK Team, “Agentic ai: A complete guide to autonomous ai agents,” https://www.videosdk.live/developer-hub/ai_agent/agentic-ai, 2024, accessed: 2025-06-01

  48. [56]

    Auto-gpt performance: Issue #5190,

    Significant Gravitas Contributors, “Auto-gpt performance: Issue #5190,” https://github.com/Significant-Gravitas/AutoGPT/issues/5190, 2023, gitHub Issue; Accessed: 2025-06-01

  49. [57]

    Monotonic value function factorisation for deep multi- agent reinforcement learning,

    T. Rashid, M. Samvelyan, C. S. De Witt, G. Farquhar, J. Foerster, and S. Whiteson, “Monotonic value function factorisation for deep multi- agent reinforcement learning,”Journal of Machine Learning Research, vol. 21, no. 178, pp. 1–51, 2020

  50. [58]

    Graph-based multi- agent reinforcement learning for large-scale uavs swarm system control,

    B. Zhao, M. Huo, Z. Li, Z. Yu, and N. Qi, “Graph-based multi- agent reinforcement learning for large-scale uavs swarm system control,” Aerospace Science and Technology, vol. 150, p. 109166, 2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.