Pith. sign in

REVIEW 3 major objections 5 minor 21 references

AI Agent Communication from Internet Architecture Perspective: Challenges and Opportunities

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Five Internet-era design pillars can organize AI agent communication

desk verdict A usable, honest framing paper on agent communication that overclaims its five-pillar 'foundational' status; the survey is competent but the DNS-to-capability-discovery analogy is the softest load-bearing spot. read the letter →

arxiv 2509.02317 v1 pith:YIFLP3BC submitted 2025-09-02 cs.NI

Chenguang Du , Chuyi Wang , Yihan Chao , Xiaohui Xie , Yong Cui This is my paper · ORCID
classification cs.NI
keywords AIagentcommunicationmulti-agentsystemsInternetarchitectureprotocolstandardizationdiscoverysecurityqualityofserviceinteroperability
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

This paper argues that the fragmented sprawl of AI agent communication protocols can be understood and guided using architectural lessons from the Internet. It distills those lessons into five pillars—scalability, security, real-time performance, high performance, and manageability—and claims these are the foundational issues for a future 'Internet of Agents.' For each pillar, it maps current bottlenecks (e.g., quadratic communication growth, semantic discovery gaps, delegation chains, latency vs. structure tension, heterogeneous traffic, LLM integration) to concrete opportunities (agent identity, capability discovery, autonomous-agent clusters, capability authentication, minimal context sharing, reliability-tunable transport, hierarchical QoS, semantic-aware manageability). The paper is a position/review contribution, not an experiment; its central claim is that a protocol ecosystem designed around these five axes will be more scalable, secure, and sustainable than today's siloed designs.

What carries the argument

The organizing device is the five-pillar taxonomy: scalability, security, real-time performance, high performance, and manageability, distilled from Internet architecture and applied section-by-section to agent communication. Each pillar is paired with an Internet mechanism—packet switching/regional autonomy for scalability, OAuth/TLS/PKI for security, UDP/QUIC/WebRTC/CDN for real-time, CDN/load balancing/QoS for high performance, and SNMP/NETCONF/observability stacks for manageability—to generate corresponding agent-side opportunities. The taxonomy does the argumentative work: it converts a scattered protocol survey into a structured bottleneck analysis and a standardization agenda.

What would settle it

A large-scale deployment of 100,000-plus agents across independent organizations that achieves trusted collaboration through flat, full-mesh discovery, no standardized identity hierarchy, and no QoS classes would undermine the claim that the five pillars are foundational. A weaker test: a historical case showing a pre-LLM agent framework failed because of missing incentives, not missing scalability, security, or manageability, which would weaken the taxonomy's diagnostic power.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is a taxonomy: after reviewing agent communications from KQML and FIPA to MCP, A2A, and ANP, the authors claim that recurring failures cluster along five dimensions inherited from Internet architecture. They call these 'foundational issues that will determine the future viability of an interconnected Internet of Agents.' For each element, the paper extracts an Internet mechanism and translates it into an agent-specific opportunity. The unifying assertion is that no single protocol can solve interoperability; like the Internet, agent communication needs architectural composition—layered naming, delegated identity, tunable transport, QoS/QoE mapping, and

Load-bearing premise

The paper assumes that Internet-era design answers—packet-switched hierarchy, DNS-style naming, layered security, QoS, and manageability—transfer to an agent world where discovery is semantic and interactions carry delegation and context; if those differences are fundamental rather than surface-level, the five pillars could point standardization in the wrong direction.

Editorial extensions

If this is right

  • Standardization should proceed at two layers at once: a common transport substrate built on modern QUIC-class protocols, plus an application layer for discovery, semantics, and negotiation, rather than betting on only one end of the stack.
  • Agent discovery will need to become a capability-oriented, hierarchical service—a 'DNS for capabilities'—with caching and aggregation, instead of exact-name resolution.
  • Security standards must handle cascading delegation from user to agent to agent, plus cryptographically verifiable capability attestation, not just endpoint authentication.
  • Real-time agent control and high-volume data exchange need reliability-tunable protocols that select delivery guarantees by message criticality, from fully reliable emergency stops to loss-tolerant sensor updates.
  • A manageable agent ecosystem requires explicit QoS classes mapped to agent-level QoE metrics, plus governance mechanisms against manipulation such as 'Agent SEO' and reputation gaming.

Reading between the lines

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

  • The paper leaves untested whether the five-pillar taxonomy predicts protocol adoption; an empirical scoring of existing frameworks against the five dimensions could show which pillars actually determine cross-organization traction.
  • If the analogy holds, the decisive competition will shift from protocol syntax to who controls agent discovery and identity—the 'entry points' the paper discusses—making reputation and ranking systems as important to agents as PageRank was to the web.
  • One can read the historical record as the paper's own implicit falsification test: KQML and FIPA had rich semantics and still failed, suggesting that semantic expressiveness without a scalable discovery and governance substrate was the fatal gap. The taxonomy explains that retroactively, but only future deployments can confirm it forward.
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

3 major / 5 minor

Summary. This paper argues that AI agent communication is fragmented and would benefit from a systematic perspective derived from Internet architecture. It identifies five key elements—scalability, security, real-time performance, high performance, and manageability—and claims these capture the fundamental challenges of agent communication. The paper reviews representative technologies from the pre-LLM and LLM eras (KQML/FIPA-ACL, OAuth, DID, MCP, A2A, ANP, etc.), analyzes each element's challenges and opportunities, and discusses standardization at both transport and application layers, including strategic competition for entry points. It concludes by acknowledging limitations regarding coverage, deployment validation, and standardization impact.

Significance. If the proposed taxonomy is accepted as a useful abstraction, the paper provides a valuable bridging of mature Internet architecture principles with the nascent agent-communication ecosystem. Its strengths include a broad and current protocol survey (Table I), concrete design proposals (e.g., Global Agent Identity System, Capability Discovery System, reliability-tunable protocols, hierarchical QoS), and attention to ongoing standardization efforts at W3C, ISO, and IETF. The paper makes no machine-checked or empirical claims, but it does offer a falsifiable organizational framework that could guide future protocol design. However, the central contribution—the five-element taxonomy—is asserted rather than derived or validated, and one load-bearing recommendation is internally strained. The significance therefore rests on whether the framework is subsequently confirmed by expert elicitation, case studies, or comparative analysis.

major comments (3)
  1. [Section III.A] The central claim that the five elements 'capture the fundamental challenges' is asserted, not demonstrated. No methodology is given for how these five elements were distilled from Internet evolution, no comparison is made with alternative taxonomies (e.g., existing multi-agent communication surveys, distributed-systems quality attributes), and no empirical or expert-elicitation support is provided. The paper's own Section V concedes that practical deployment cases remain limited and standardization impacts are unvalidated. This is load-bearing because the entire analysis is organized around this taxonomy. The authors should either reframe the paper as proposing a perspective/hypothesis needing validation, or add evidence such as a systematic literature corpus, expert validation, or a comparison with existing frameworks.
  2. [Section III.B.2 vs. III.B.3] The scalability pillar contains an internal inconsistency that affects the main prescription. Section III.B.2 states that agent discovery 'departs fundamentally from the Internet's DNS' and is 'similar to a search engine, matching user intent to ranked capabilities through semantic resolution.' Yet Section III.B.3 proposes a Capability Discovery System that 'would function as a DNS for the agent era,' relying on 'caching and hierarchical aggregation' for scale. The paper never shows that semantic capability queries have the locality, cacheability, and bounded aggregation properties that make DNS and BGP scalable. Without such an argument, the analogy is unsupported. Please either remove the DNS framing, or provide a concrete analysis of why semantic discovery queries can be made scalable through caching/hierarchical aggregation despite their search-like nature.
  3. [Section III.A and III.D/III.E] The taxonomy's internal boundaries are not crisply defined. In particular, 'real-time performance' and 'high performance' are treated as separate elements, but the discussion overlaps substantially: CDN, edge computing, and latency appear in both; 'real-time' is defined via latency and consistency, while 'high performance' includes bandwidth, concurrency, and throughput, yet Section III.E.2's 'heterogeneous scheduling complexity' mixes latency and reliability with throughput. This weakens the claim that the five elements are distinct key dimensions. The authors should clarify the distinguishing criteria and, if the categories are intentionally overlapping, state so and explain why the overlap does not undermine the taxonomy's utility.
minor comments (5)
  1. [References] Reference [7] cites a 1993 paper titled 'GRPC: A communication cooperation mechanism in distributed systems' as the source for gRPC. This appears to be a different, earlier GRPC system; the modern gRPC protocol was released by Google in 2015. Please correct the citation to the actual gRPC specification or project.
  2. [Section V] The conclusion's limitation statement is useful but too brief. Move it to a dedicated 'Threats to Validity' or 'Methodology and Limitations' subsection, and expand on how the lack of empirical validation should temper the paper's prescriptive recommendations.
  3. [Section IV.A.2] The terms 'AgentDNS' and 'Internet of Agents Protocol (IoA Protocol)' are mentioned without citations. Please provide references for these systems so that readers can verify the claims.
  4. [Table I] The table's legend uses symbols that may not render clearly in all formats. Verify that the symbols for 'not supported', 'partially supported', and 'well supported' print unambiguously in the final PDF.
  5. [Section III.B.3] The term 'Autonomous Agent' for the proposal to group agents like Internet Autonomous Systems is potentially confusing, since the paper uses 'autonomous' elsewhere to describe AI agents. Consider renaming this opportunity (e.g., 'Agent Autonomous Systems') to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is a taxonomy/survey with no derivation chain, no fitted parameters, and no load-bearing self-citations.

full rationale

This is a position paper that constructs a five-element taxonomy (scalability, security, real-time performance, high performance, manageability) from Internet architecture and uses it to organize a survey of agent communication challenges and opportunities. There is no formal derivation, no predictive model, and no fitted parameter; consequently, none of the circularity patterns (self-definition, fitted-input-called-prediction, self-citation chains, imported uniqueness theorems, ansatz-smuggling, or renaming) apply. The paper explicitly acknowledges that agent discovery departs fundamentally from DNS (Section III.B.2), undermining any claim that it silently assumes the Internet analogy as a mathematical identity. The limitation paragraph in Section V states that the framework's impact 'has yet to be validated,' which is an honest admission rather than a circular justification. All cited protocols and standards are external works; no load-bearing self-citations are present. The five elements are an asserted abstraction, not an output derived from inputs, so the derivation chain is not circular.

Assumptions & free parameters 0 free parameters · 4 assumptions · 2 invented entities

The paper's central claim rests on domain assumptions about the transferability of Internet design lessons and on an ad hoc taxonomy: the five elements are asserted rather than derived. The proposed systems (GAIS, CDS) are speculative opportunities, not demonstrated artifacts.

assumptions (4)
  • domain assumption Internet architecture is a transferable source of design principles for agent communication.
    Invoked in Section III.A, where the paper adopts Internet architecture 'not merely as a historical analogy, but as a source of time-tested design elements'. If the domains differ too much, the five elements lose force.
  • ad hoc to paper The five elements (scalability, security, real-time, high performance, manageability) are the right, roughly complete set of fundamental agent-communication challenges.
    Stated in Section I and III.A. No method is given for deriving the set, and no comparison with alternative taxonomies. The paper then populates each element with challenges of its choosing.
  • domain assumption Agent ecosystems will scale to Internet-like global scope and heterogeneity.
    The framing assumes the end state is open, cross-domain, global-scale multi-agent communication (Section I and III.B). If agents remain platform-bound, the Internet lens over-weights scalability and governance.
  • domain assumption Agent discovery is fundamentally a semantic, intent-to-capability matching problem rather than a DNS-like name-address resolution.
    Section III.B.2 asserts this and builds the discovery opportunities on it; it is a modeling choice, not an established fact.
invented entities (2)
  • Global Agent Identity System (GAIS)
    purpose: Proposed hierarchical identity/addressing layer for agents, mentioned in Section III.B.3 as a development opportunity.
    Introduced as a design suggestion without implementation, specification, or empirical validation.
  • Capability Discovery System (CDS)
    purpose: Proposed distributed capability-resolution service, described as 'a DNS for the agent era' in Section III.B.3.
    Introduced as a design suggestion without implementation, specification, or empirical validation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AI Agent Communication from Internet Architecture Perspective: Challenges and Opportunities." pith.science (2026). https://pith.science/paper/YIFLP3BC

@misc{pith2026250902317,
  author       = {Pith},
  title        = {Pith review of: AI Agent Communication from Internet Architecture Perspective: Challenges and Opportunities},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YIFLP3BC}},
  note         = {Machine review of arXiv:2509.02317}
}
read the original abstract

The rapid development of AI agents leads to a surge in communication demands. Alongside this rise, a variety of frameworks and protocols emerge. While these efforts demonstrate the vitality of the field, they also highlight increasing fragmentation, with redundant innovation and siloed designs hindering cross-domain interoperability. These challenges underscore the need for a systematic perspective to guide the development of scalable, secure, and sustainable AI agent ecosystems. To address this need, this paper provides the first systematic analysis of AI agent communication from the standpoint of Internet architecture-the most successful global-scale distributed system in history. Specifically, we distill decades of Internet evolution into five key elements that are directly relevant to agent communication: scalability, security, real-time performance, high performance, and manageability. We then use these elements to examine both the opportunities and the bottlenecks in developing robust multi-agent ecosystems. Overall, this paper bridges Internet architecture and AI agent communication for the first time, providing a new lens for guiding the sustainable growth of AI agent communication ecosystems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 20 canonical work pages

  1. [7]

    Grpc: A communication cooperation mechanism in distributed systems,

    X. Wang, H. Zhao, and J. Zhu, “Grpc: A communication cooperation mechanism in distributed systems,” ACM SIGOPS Operating Systems Review, vol. 27, no. 3, pp. 75–86, 1993

    work page 1993

  2. [1]

    Agent-to-agent (a2a) project pages and key concepts,

    Google, “Agent-to-agent (a2a) project pages and key concepts,” https: //github.com/google/a2a, 2025, accessed: 2025-08-27

    work page 2025

  3. [2]

    Agent network protocol (anp) specifications and repository,

    ANP Contributors, “Agent network protocol (anp) specifications and repository,” https://github.com/agent-network-protocol/ AgentNetworkProtocol, 2025, accessed: 2025-08-27

    work page 2025

  4. [3]

    Kqml-a language and protocol for knowledge and information exchange,

    T. Finin, R. Fritzson, D. P. McKay, R. McEntire et al., “Kqml-a language and protocol for knowledge and information exchange,” in 13th Int. Distributed Artificial Intelligence Workshop , 1994, pp. 93–103

    work page 1994

  5. [4]

    Fipa acl message structure specification,

    A. Fipa, “Fipa acl message structure specification,” Foundation for Intel- ligent Physical Agents, http://www. fipa. org/specs/fipa00061/SC00061G. html (30.6. 2004) , 2002

    work page 2004

  6. [5]

    Rfc 6749: The oauth 2.0 authorization framework,

    D. Hardt, “Rfc 6749: The oauth 2.0 authorization framework,” 2012

    work page 2012

  7. [6]

    Json-ld 1.1–a json-based serialization for linked data,

    G. Kellogg, P.-A. Champin, and D. Longley, “Json-ld 1.1–a json-based serialization for linked data,” Ph.D. dissertation, W3C, 2019

    work page 2019

  8. [8]

    Web service to json-rpc transformation,

    C. Samsel, S. G ¨okay, P. Heiniz, and K.-H. Krempels, “Web service to json-rpc transformation,” in International Conference on Software Engineering and Applications , vol. 2. SCITEPRESS, 2013, pp. 214– 219

    work page 2013

Show all 21 references
  1. [9]

    Decentralized identifiers (dids) v1. 0,

    D. Reed, M. Sporny, D. Longley, C. Allen, R. Grant, M. Sabadello, and J. Holt, “Decentralized identifiers (dids) v1. 0,” Draft Community Group Report, 2020

    2020

  2. [10]

    Model context protocol (mcp): Announcement and specifi- cation,

    Anthropic, “Model context protocol (mcp): Announcement and specifi- cation,” https://github.com/modelcontextprotocol, 2025, accessed: 2025- 08-27

    2025

  3. [11]

    Agent name service (ans): A universal directory for secure ai agent discovery and interoperability,

    K. Huang, V . S. Narajala, I. Habler, and A. Sheriff, “Agent name service (ans): A universal directory for secure ai agent discovery and interoperability,” arXiv preprint arXiv:2505.10609 , 2025

    2025 arXiv

  4. [12]

    Developing multi-agent systems with jade,

    F. Bellifemine, A. Poggi, and G. Rimassa, “Developing multi-agent systems with jade,” in International workshop on agent theories, ar- chitectures, and languages . Springer, 2000, pp. 89–103

    2000

  5. [13]

    Autogen: Enabling next-gen llm applications via multi-agent conversations,

    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 conversations,” in First Conference on Language Modeling , 2024

    2024

  6. [14]

    Dify documentation: Agent node and workflow orchestra- tion,

    Dify Team, “Dify documentation: Agent node and workflow orchestra- tion,” https://docs.dify.ai/, 2025, accessed: 2025-08-27

    2025

  7. [15]

    Langgraph studio and documentation,

    LangChain and Langgraph, “Langgraph studio and documentation,” https://www.langchain.com/langgraph, 2025, accessed: 2025-08-27

    2025

  8. [16]

    Openai agents sdk documentation and repository,

    OpenAI, “Openai agents sdk documentation and repository,” https: //openai.github.io/openai-agents-python/, 2025, accessed: 2025-08-27

    2025

  9. [17]

    Energy management agent functional requirements and interfaces,

    ISO/IEC JTC 1/SC 25, “Energy management agent functional requirements and interfaces,” International Organization for Standardization / International Electrotechnical Commission, ISO/IEC Standard 15067-3-30:2024, March 2024. [Online]. Available: https://www.iso.org/standard/89221.html

    2024

  10. [18]

    Protocol of energy management agents for demand-response energy management and interactions among these agents,

    ——, “Protocol of energy management agents for demand-response energy management and interactions among these agents,” International Organization for Standardization / International Electrotechnical Commission, ISO/IEC Standard 15067-3-31:2024, March 2024. [Online]. Available: ...

    2024

  11. [19]

    Agent name service (ans): A universal directory for secure ai agent discovery and interoperability,

    H. Ken, S. N. Vineeth, H. Idan, and S. Akram, “Agent name service (ans): A universal directory for secure ai agent discovery and interoperability,” Internet Engineering Task Force, Internet- Draft draft-narajala-ans-00, May 2025. [Online]. Available: https: //datatracker.ietf....

    2025

  12. [20]

    Aauth - agentic authorization oauth 2.1 extension,

    R. Jonathan and W. Pat, “Aauth - agentic authorization oauth 2.1 extension,” Internet Engineering Task Force, Internet-Draft draft-rosenberg-oauth-aauth-00, July 2025. [Online]. Available: https: //datatracker.ietf.org/doc/draft-rosenberg-oauth-aauth/

    2025

  13. [21]

    Digital identity management for ai agent communication protocols,

    Y . Kehan and L. Peng, “Digital identity management for ai agent communication protocols,” Internet Engineering Task Force, Internet- Draft draft-yl-agent-id-requirements-00, July 2025. [Online]. Available: https://datatracker.ietf.org/doc/draft-yl-agent-id-requirements/

    2025

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.