acceptodds
Withdrawn from ICLR 2027

NegoLat: Negotiating What to Communicate in Multi-Agent Systems

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

A central yet under-explored question in multi-agent systems (MAS) is what to communicate. Existing protocols, whether textual or latent, remain sender-centric: an upstream agent composes one message from its own context and delivers the same payload to every downstream agent, which participates only after the message has been formed. Such one-shot communication mismatches receivers that need different information and leaves incomplete or conflicting messages unrepaired. We propose NegoLat, a training-free receiver-negotiated latent communication paradigm. On each edge, the sender first exposes a semantic probe; the downstream receiver audits it against its own role and selectively requests refinement; the sender then regenerates a receiver-specific message only for those who asked. We further analyze when receiver-guided regeneration is beneficial and how a bounded request budget affects information coverage. Across ten benchmarks spanning mathematical reasoning, code generation, and visual question answering, NegoLat consistently outperforms text-based and fixed latent communication, improving over the best baseline by pp (Qwen3-4B-Instruct) and pp (Qwen3-VL-4B) on average under Hierarchical MAS, with limited overhead. Communication analyses show that NegoLat adaptively adjusts transmitted information according to downstream requirements. Code will be released.

Primary area:
foundation or frontier models, including LLMs
Keywords:
Multi-Agent Systems, Receiver-Negotiated Communication

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_99RtOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper proposes NegoLat, a training-free communication protocol that allows downstream agents to inspect upstream messages and request receiver-specific revisions before latent transfer. Receivers either accept a textual semantic probe or issue structured requests identifying missing, ambiguous, or conflicting information. The sender selectively regenerates the message, and the resulting hidden states are transferred through the existing StateBridge interface. Experiments on ten benchmarks covering different tasks report generally improved performance under sequential and hierarchical agent structures, with ablations supporting the main negotiation components.

Critical strengths
  1. The studied problem is well motivated and valuable. Allowing receivers to identify missing or conflicting information before using an upstream message is a natural and reasonable approach to improving agent collaboration.
  2. The proposed protocol is simple, training-free, and interpretable. Structured requests and selective regeneration provide a practical way to tailor messages to different receivers.
  3. The experiments cover three task families, four backbones, and two communication structures. Ablations of receiver roles, request structure, additional generation, shared payloads, and joint auditing provide useful evidence for the proposed design.
Critical weaknesses
  1. I feel that the methodological novelty and comparative advantage are relatively limited. The latent transfer mechanism is inherited from StateBridge, making receiver-guided feedback the main contribution. However, comparisons are limited to Single, TextMAS, and StateBridge, without testing text-based negotiation or direct receiver-role conditioning without an audit. The unconditioned two-pass ablation is useful, but does not establish an advantage over stronger feedback-based refinement. Comparisons with communication optimization methods such as AgentPrune and G-Designer, under comparable inference budgets and with optimization costs reported separately, would further clarify the performance–cost advantage.

  2. The theoretical analysis provides very limited insight into the proposed method. Theorem 1 is a direct expected-value decomposition, while Theorem 2 assumes independent recovery opportunities and preservation of recovered information, which are not established for complete message regeneration. Moreover, Appendix C.4 measures defects mentioned in requests rather than defects actually repaired. The practical justification for the request budget therefore remains limited.

  3. It seems that the claim of limited overhead is not sufficiently supported by the runtime results. In Table 5, sequential MBPP+ takes 149.97 seconds with NegoLat, compared with 38.52 seconds for StateBridge and 54.62 seconds for TextMAS. Short feedback and selective regeneration do not eliminate auditing costs. The authors are suggested to qualify the efficiency claim and compare performance under matched end-to-end inference budgets.

  4. The robustness and generality of the improvements require stronger evidence. All evaluated backbones have only 3B–4B parameters (especially for a training-free method), with agents sharing the same backbone and using manually designed task vocabularies. Several gains are small, and some settings show regressions. Although selected experiments include repeated runs, the main results lack uncertainty estimates, making the reliability of these improvements difficult to assess.

Questions for authors
  1. How does NegoLat compare with text-based negotiation, directly conditioning the sender on the receiver’s role, or feedback-based refinement, given the same inference budget?
  2. Could you check whether the defects identified in requests are actually fixed after regeneration, and whether new errors are introduced? Does the request-budget finding also hold for naturally occurring errors?
  3. Could you report uncertainty estimates for the main results and test a larger backbone? It would also help to explain where the additional latency in Table 5 comes from.
Reviewer_D3gPOpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
3
Soundness
3
Read review
Paper summary

The paper proposes to let receiver negotiate what to communicate in latent communication of multi agent LLM systems. essentially the receiver is given a way to trigger regeneration of the message if the receiver deem it not good.

Critical strengths
  1. The paper introduces ACK/REQUEST protocol when designing the negotiation procedure, which is an interesting bridging from the network domain, though with limited depth.

  2. With informative ablation studies.

Critical weaknesses
  1. Novelty is significantly overclaimed. For example, see "AgentAsk: Multi-Agent Systems Need to Ask".

  2. Accuracy gains are modest and did not control for additional inference computation. What if the agent reasons while negotiating? The performance gain is not clearly isolated.

Questions for authors

N/A

Reviewer_SdXHOpenReview ↗
Rating
1
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

The paper proposes NegoLat, a training-free protocol for receiver-negotiated latent communication in multi-agent systems. A sender exposes a textual semantic probe, and the receiver either accepts it or issues structured requests for refinement. Accepted messages reuse cached states; requested messages are regenerated for the receiver. The selected hidden states are then transferred through the StateBridge alignment interface. Experiments cover four open-weight backbones and ten benchmarks spanning mathematical reasoning, code generation, and visual question answering under sequential and hierarchical topologies.

Critical strengths
  1. Explicit protocol without additional training. The ACK/REQUEST mechanism, bounded request schema, and reuse of cached states are clearly specified. The protocol requires no parameter updates, making its communication procedure straightforward to inspect.
  2. Evaluation across several task types. The experiments include mathematical reasoning, code generation, and VQA under two topologies. Ablations also examine receiver-role conditioning, structured requests, and receiver-specific messages.
Critical weaknesses
  1. Limited methodological novelty. Allowing receivers to request changes to sender messages resembles the iterative feedback used in existing multi-agent systems, such as multi-agent debate (MAD), where agents revise their answers through discussion. The paper should more clearly establish what is fundamentally new beyond adapting this feedback pattern to latent communication.
  2. Modest gains relative to latency. The accuracy improvements in the main experiments appear modest relative to the substantial latency increases reported in Table 5. A clearer justification of this accuracy–latency trade-off is needed.
  3. Limited backbone coverage. Although the paper evaluates four open-weight backbones, they are all relatively small models. Experiments with additional model families and larger models would strengthen the evidence for the method’s generality.
  4. Limited evidence on naturally occurring communication errors. Figure 3 measures error detection and recovery using artificially corrupted probes. This does not establish that the method prevents errors from arising during ordinary communication. Further analysis of naturally occurring message errors would help clarify how effectively the protocol limits their propagation in practice.
Questions for authors

Just see weeknesses

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.