pith. sign in

arxiv: 2606.19116 · v1 · pith:3526A2ROnew · submitted 2026-06-17 · 💻 cs.AI · cs.CY

Towards an Agent-First Web: Redesigning the Web for AI Agents

Eranga Bandara , Ross Gore , Ravi Mukkamala , Asanga Gunaratna , Safdar H. Bouk , Xueping Liang , Peter Foytik , Abdul Rahman
show 13 more authors
Sachini Rajapakse Isurunima Kularathna Pramoda Karunarathna Chalani Rajapakse Ng Wee Keong Kasun De Zoysa Tharaka Hewa Amin Hass Wathsala Herath Aruna Withanage Nilaan Loganathan Atmaram Yarlagadda Sachin Shetty
This is my paper

Pith reviewed 2026-06-26 20:47 UTC · model grok-4.3

classification 💻 cs.AI cs.CY
keywords AI agentsweb redesignagent-first internetATMLepistemic recursionweb accesscontent provenancehuman supervision
0
0 comments X

The pith

The web's human-centric design must be replaced by ten principles that make AI agents first-class participants across access, economics, and content layers.

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

The paper argues that the web was built on the assumption that humans are its primary consumers, an assumption now invalidated by AI agents acting as intermediaries. This mismatch produces blocking, exclusion, and economic friction, so the authors propose changes to access rights, payment models, and content supervision. A sympathetic reader would care because the alternative is a web where agents are treated as threats and content drifts away from human origins through repeated AI consumption and generation.

Core claim

The rapid emergence of AI agents as intermediaries between humans and web content invalidates the human-centric assumption of the web, requiring a principled redesign across access, economic, and content layers that constitutes ten design principles for an agent-first internet.

What carries the argument

the three-layer architecture (access, economic, content) together with the Agent Text Markup Language (ATML) four-level human supervision model and cryptographic provenance chain

If this is right

  • Agents acting for humans inherit equivalent access rights governed by rate limiting and agent identification metadata in HTTP requests.
  • An agent's economic obligation mirrors that of the human it represents through intent-based tiers and token-based metering instead of pageviews.
  • The four-level ATML model with cryptographic provenance counters epistemic recursion by anchoring AI-generated content to human supervision.
  • A commissioned content economy ties AI production to human intentionality rather than autonomous loops.
  • Dual-layer architecture serves both human-readable and agent-optimized content from the same domain.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Standards bodies would need to formalize agent metadata headers as a required extension to existing HTTP practices.
  • Content platforms might develop separate APIs optimized for agent consumption while keeping the human-facing version unchanged.
  • Legal systems could treat agent actions as direct proxies for human intent, affecting liability and contract enforcement online.
  • Long-term testing of ATML on live sites could measure whether the supervision tiers actually reduce content drift over multiple generations.

Load-bearing premise

The proposed mechanisms such as dual-layer content serving, intent-based economic tiers, and the ATML supervision model can be implemented at scale by existing web operators and standards bodies without major technical, legal, or adoption barriers.

What would settle it

Widespread deployment of the proposed agent identification headers, dual-layer serving, and ATML markup on major sites followed by continued blanket blocking or measurable detachment of web content from human ground truth would show the redesign fails to solve the problem.

Figures

Figures reproduced from arXiv: 2606.19116 by Abdul Rahman, Amin Hass, Aruna Withanage, Asanga Gunaratna, Atmaram Yarlagadda, Chalani Rajapakse, Eranga Bandara, Isurunima Kularathna, Kasun De Zoysa, Ng Wee Keong, Nilaan Loganathan, Peter Foytik, Pramoda Karunarathna, Ravi Mukkamala, Ross Gore, Sachini Rajapakse, Sachin Shetty, Safdar H. Bouk, Tharaka Hewa, Wathsala Herath, Xueping Liang.

Figure 1
Figure 1. Figure 1: The broken economic loop. In the human web (top), content consumption [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The epistemic recursion loop. Human-origin knowledge enters the web through [PITH_FULL_IMAGE:figures/full_fig_p016_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Three-layer failure coupling. The access, economic, and content layer failures of [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Agent identification metadata flow. An AI agent sends standardized metadata [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Graduated rate limiting model replacing blanket blocking. Agent access is tiered [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Dual-layer web architecture and migration path. The same domain serves HTML [PITH_FULL_IMAGE:figures/full_fig_p025_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Intent-based economic tier model. Agent content access behaviors are classified [PITH_FULL_IMAGE:figures/full_fig_p029_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Token-based subscription model. Content access is metered in tokens consumed [PITH_FULL_IMAGE:figures/full_fig_p032_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The commissioned content economy. On the supply side, a human entity com [PITH_FULL_IMAGE:figures/full_fig_p034_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: ATML structure compared to current HTML. HTML carries significant visual [PITH_FULL_IMAGE:figures/full_fig_p039_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Human supervision tier model. Content is classified into four levels based [PITH_FULL_IMAGE:figures/full_fig_p042_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Agent content discoverability architecture. An agent queries an agent search [PITH_FULL_IMAGE:figures/full_fig_p045_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The web’s social contract: original versus agent-first. The original contract [PITH_FULL_IMAGE:figures/full_fig_p049_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: The unified agent-first web framework. The agent-as-human-proxy principle [PITH_FULL_IMAGE:figures/full_fig_p052_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Open challenges in the agent-first web. Six challenge areas remain unresolved [PITH_FULL_IMAGE:figures/full_fig_p056_15.png] view at source ↗
read the original abstract

The World Wide Web was built on an assumption held for three decades: the primary consumer of web content is a human being. This permeates every layer; its access model presumes human visitors, its economics rest on human attention, and its content targets human perception. The rapid emergence of AI agents as intermediaries between humans and web content invalidates this assumption. Yet the web resists agents through blanket blocking, CAPTCHA-based exclusion, and economic models that treat agent access as extraction rather than legitimate interaction. This paper proposes a principled redesign across three layers. At the access layer, agents acting for humans should inherit equivalent access rights, governed by rate limiting and agent identification metadata in HTTP requests, analogous to browser headers, alongside a dual-layer architecture serving human-readable and agent-optimized content from the same domain. At the economic layer, we propose an intent-based tier framework grounded in the agent-as-human-proxy principle: an agent's economic obligation mirrors that of the human it represents. A token-based subscription model meters content in tokens rather than pageviews, alongside a commissioned content economy anchoring AI content production in human intentionality. At the content layer, we identify epistemic recursion, the self-referential loop in which AI-generated content is consumed by agents to produce further content, progressively detaching web knowledge from human ground truth. We propose the Agent Text Markup Language (ATML), a four-level human supervision tier model, and a cryptographic provenance chain to counter this threat. Together these constitute ten design principles for an agent-first internet, one in which agents are first-class citizens whose integration requires renegotiating the web's foundational social contract across access, economics, and content.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper claims that the human-centric assumptions of the World Wide Web are invalidated by AI agents acting as intermediaries, resulting in blocking and mismatched economics; it proposes a three-layer redesign (access, economic, content) consisting of ten design principles, including HTTP agent metadata and dual-layer serving, intent-based economic tiers with token metering and commissioned content, and the Agent Text Markup Language (ATML) with four-level human supervision plus cryptographic provenance to address epistemic recursion.

Significance. If the proposals prove viable, the work could supply a coherent conceptual framework for integrating AI agents into web infrastructure while mitigating risks of access exclusion and knowledge detachment from human sources. The manuscript identifies timely problems and offers structured principles, but its significance is limited by the absence of any feasibility analysis, implementation details, or validation.

major comments (3)
  1. [Abstract] Abstract and three-layer architecture description: the claim that the proposed dual-layer content serving, intent-based tier framework, and ATML four-level model constitute realizable solutions at web scale rests on the unexamined premise that operators and standards bodies will voluntarily adopt and enforce them; no analysis of incentive alignment, migration costs, legal barriers, or historical precedents for cross-layer standardization is supplied, which is load-bearing for the central assertion that these changes renegotiate the web's social contract.
  2. [Content layer] Content layer proposal: the ATML four-level supervision model and cryptographic provenance chain are presented as countermeasures to epistemic recursion, yet the manuscript supplies neither a formal specification of the levels, example markup, nor any demonstration (even conceptual) that the mechanism would break self-referential AI-content loops in practice.
  3. [Economic layer] Economic layer proposal: the intent-based tier framework and token-based subscription model are asserted to mirror human obligations and anchor content in human intentionality, but no discussion of integration with existing payment rails, enforcement against agent abuse, or economic viability under current pageview-driven incentives is included.
minor comments (2)
  1. [Abstract] The acronym ATML is introduced without expansion on first use or comparison to existing markup standards such as HTML or XML extensions.
  2. The ten design principles are enumerated in the abstract but not explicitly mapped or numbered in the body, making it difficult to track which mechanisms correspond to which principle.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the timeliness of the identified problems. The manuscript presents a conceptual framework of ten design principles rather than an implemented system or economic analysis. We respond point-by-point to the major comments below, maintaining that the contribution lies in the structured identification of mismatches and the proposed principles.

read point-by-point responses

  1. Referee: [Abstract] Abstract and three-layer architecture description: the claim that the proposed dual-layer content serving, intent-based tier framework, and ATML four-level model constitute realizable solutions at web scale rests on the unexamined premise that operators and standards bodies will voluntarily adopt and enforce them; no analysis of incentive alignment, migration costs, legal barriers, or historical precedents for cross-layer standardization is supplied, which is load-bearing for the central assertion that these changes renegotiate the web's social contract.

    Authors: The manuscript does not assert that the proposals are immediately realizable at web scale or that voluntary adoption is assured. It argues that the human-centric assumptions are invalidated and that the listed principles provide a coherent framework for renegotiating the social contract. The central claim concerns the necessity of redesign, not its guaranteed enforcement. Detailed incentive or migration analysis lies outside the scope of a principles paper; analogous shifts (e.g., HTTPS adoption) occurred through iterative standards processes without initial exhaustive feasibility studies in the originating documents. revision: no

  2. Referee: [Content layer] Content layer proposal: the ATML four-level supervision model and cryptographic provenance chain are presented as countermeasures to epistemic recursion, yet the manuscript supplies neither a formal specification of the levels, example markup, nor any demonstration (even conceptual) that the mechanism would break self-referential AI-content loops in practice.

    Authors: ATML and the provenance chain are introduced at the conceptual level to illustrate how human supervision tiers and cryptographic anchoring can interrupt epistemic recursion by enabling traceability to human-grounded sources. A full formal specification and implementation demonstration would require a follow-on technical paper. The conceptual mechanism is that agents can query supervision level and provenance metadata to prioritize or filter content, thereby breaking pure self-referential loops. We will add a short illustrative ATML markup example in revision. revision: partial

  3. Referee: [Economic layer] Economic layer proposal: the intent-based tier framework and token-based subscription model are asserted to mirror human obligations and anchor content in human intentionality, but no discussion of integration with existing payment rails, enforcement against agent abuse, or economic viability under current pageview-driven incentives is included.

    Authors: The economic principles rest on the agent-as-proxy premise, shifting from pageview to token metering and intent-based tiers to align obligations with represented humans. Practical integration with payment rails, abuse enforcement, and transition from existing incentives are implementation details that follow from the conceptual model rather than prerequisites for stating the principles. The paper highlights the mismatch with current models but does not claim immediate economic viability. revision: no

Circularity Check

0 steps flagged

No circularity: design proposal with no derivations or self-referential reductions

full rationale

The paper is a forward-looking conceptual proposal outlining ten design principles for an agent-first web across access, economic, and content layers. It contains no equations, fitted parameters, predictions derived from data, or logical derivations that reduce to their own inputs. The central claims rest on reasoning about observed web trends and proposed mechanisms (dual-layer serving, ATML tiers, cryptographic provenance) without any self-citation load-bearing steps, uniqueness theorems, or ansatzes smuggled via prior work. No load-bearing step reduces by construction to a fit or self-reference; the argument is self-contained as normative design advice rather than a closed deductive system.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

The proposal introduces several new concepts and domain assumptions about agent behavior and content dynamics without independent evidence or prior validation in the provided abstract.

axioms (2)
  • domain assumption AI agents act as human proxies whose access rights and economic obligations should mirror those of the represented humans.
    This assumption directly supports the access and economic layer proposals in the abstract.
  • domain assumption Epistemic recursion from AI-generated content poses a growing threat that requires new markup and provenance mechanisms.
    This underpins the content layer interventions including ATML.
invented entities (2)
  • Agent Text Markup Language (ATML) no independent evidence
    purpose: Provide four-level human supervision tiers and structure content to counter epistemic recursion.
    Newly proposed markup language introduced without reference to prior implementations or testing.
  • intent-based tier framework no independent evidence
    purpose: Align agent economic obligations with human proxies using token-based metering.
    New economic model proposed in the paper without supporting analysis.

pith-pipeline@v0.9.1-grok · 5939 in / 1553 out tokens · 41717 ms · 2026-06-26T20:47:55.250060+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

62 extracted references · 1 canonical work pages

  1. [1]

    CAPTCHA: Using hard AI problems for security, in: Advances in Cryptology – EU- ROCRYPT 2003, pp

    von Ahn, L., Blum, M., Hopper, N.J., Langford, J., 2003. CAPTCHA: Using hard AI problems for security, in: Advances in Cryptology – EU- ROCRYPT 2003, pp. 294–311

    2003

  2. [2]

    Publisher revenue sharing program.https://www

    AI, P., 2024. Publisher revenue sharing program.https://www. perplexity.ai. 63

    2024

  3. [3]

    DomainKeys identified mail (DKIM) signatures – RFC 4871

    Allman, E., Callas, J., Delany, M., Libbey, M., Fenton, J., Thomas, M., 2007. DomainKeys identified mail (DKIM) signatures – RFC 4871. https://tools.ietf.org/html/rfc4871

    2007

  4. [4]

    Agent network protocol (ANP).https: //agent-network-protocol.com

    ANP Working Group, 2025. Agent network protocol (ANP).https: //agent-network-protocol.com

    2025

  5. [5]

    Model context protocol (MCP).https:// modelcontextprotocol.io

    Anthropic, 2024. Model context protocol (MCP).https:// modelcontextprotocol.io

    2024

  6. [6]

    Language models enable simple systems for gen- erating structured views of heterogeneous data lakes

    Arora, S., Yang, B., Eyuboglu, S., Narayan, A., Hojel, A., Trummer, I., R´ e, C., 2023. Language models enable simple systems for gen- erating structured views of heterogeneous data lakes. arXiv preprint arXiv:2304.09433

    arXiv 2023

  7. [7]

    Slice-mcp — fine-tuned llama-4 llm and mcp enabled 5g network slice orchestration platform, in: MILCOM 2025 - 2025 IEEE Military Communications Conference (MILCOM), pp

    Bandara, E., Bouk, S.H., Mukkamala, R., Rahman, A., Liang, X., Gore, R., Shetty, S., 2025a. Slice-mcp — fine-tuned llama-4 llm and mcp enabled 5g network slice orchestration platform, in: MILCOM 2025 - 2025 IEEE Military Communications Conference (MILCOM), pp. 1506–

    2025

  8. [8]

    doi:10.1109/MILCOM64451.2025.11310417

    work page doi:10.1109/milcom64451.2025.11310417 2025

  9. [9]

    A practical guide for designing, developing, and deploying production- grade agentic ai workflows

    Bandara, E., Gore, R., Foytik, P., Shetty, S., Mukkamala, R., Rah- man, A., Liang, X., Bouk, S.H., Hass, A., Rajapakse, S., et al., 2025b. A practical guide for designing, developing, and deploying production- grade agentic ai workflows. arXiv preprint arXiv:2512.08769

    arXiv

  10. [10]

    Think before you act–a neurocognitive governance model for autonomous ai agents

    Bandara, E., Gore, R., Gunaratna, A., Rajapakse, S., Kularathna, I., Mukkamala, R., Shetty, S., Liang, X., Hass, A., Hewa, T., et al., 2026a. Think before you act–a neurocognitive governance model for autonomous ai agents. arXiv preprint arXiv:2604.25684

    Pith/arXiv arXiv

  11. [11]

    Agentsway–software development methodology for ai agents-based teams

    Bandara, E., Gore, R., Liang, X., Rajapakse, S., Kularathne, I., Karunarathna, P., Foytik, P., Shetty, S., Mukkamala, R., Rahman, A., et al., 2025c. Agentsway–software development methodology for ai agents-based teams. arXiv preprint arXiv:2510.23664

    arXiv

  12. [12]

    Standardization of neuromuscular reflex analysis– role of fine-tuned vision-language model consortium and openai gpt- oss reasoning llm enabled decision support system

    Bandara, E., Gore, R., Shetty, S., Mukkamala, R., Rhea, C., Yarla- gadda, A., Kaushik, S., De Silva, L., Maznychenko, A., Sokolowska, 64 I., et al., 2025d. Standardization of neuromuscular reflex analysis– role of fine-tuned vision-language model consortium and openai gpt- oss reasoning llm enabled decision support system. arXiv preprint arXiv:2508.12473

    arXiv

  13. [13]

    A practical guide to agentic ai transition in organizations

    Bandara, E., Gore, R., Shetty, S., Rajapakse, S., Kularathna, I., Karunarathna, P., Mukkamala, R., Foytik, P., Bouk, S.H., Rahman, A., et al., 2026b. A practical guide to agentic ai transition in organizations. arXiv preprint arXiv:2602.10122

    arXiv

  14. [14]

    Ai trust os–a continuous governance framework for autonomous ai observability and zero-trust compliance in enterprise environments

    Bandara, E., Gunaratna, A., Gore, R., Rahman, A., Mukkamala, R., Shetty, S., Rajapakse, S., Kularathna, I., Foytik, P., Bouk, S.H., et al., 2026c. Ai trust os–a continuous governance framework for autonomous ai observability and zero-trust compliance in enterprise environments. arXiv preprint arXiv:2604.04749

    Pith/arXiv arXiv

  15. [15]

    Astride: A security threat modeling platform for agentic-ai applications

    Bandara, E., Hass, A., Gore, R., Shetty, S., Mukkamala, R., Bouk, S.H., Liang, X., Keong, N.W., De Zoysa, K., Withanage, A., et al., 2025e. Astride: A security threat modeling platform for agentic-ai applications. arXiv preprint arXiv:2512.04785

    arXiv

  16. [16]

    Towards responsible and explainable ai agents with consensus-driven reasoning

    Bandara, E., Hewa, T., Gore, R., Shetty, S., Mukkamala, R., Foytik, P., Rahman, A., Bouk, S.H., Liang, X., Hass, A., et al., 2025f. Towards responsible and explainable ai agents with consensus-driven reasoning. arXiv preprint arXiv:2512.21699

    arXiv

  17. [17]

    Model context contracts-mcp-enabled framework to integrate llms with blockchain smart contracts

    Bandara, E., Shetty, S., Mukkamala, R., Gore, R., Foytik, P., Bouk, S.H., Rahman, A., Liang, X., Keong, N.W., De Zoysa, K., et al., 2025g. Model context contracts-mcp-enabled framework to integrate llms with blockchain smart contracts. arXiv preprint arXiv:2510.19856

    arXiv

  18. [18]

    On the dangers of stochastic parrots: Can language models be too big?, in: Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT), pp

    Bender, E.M., Gebru, T., McMillan-Major, A., Shmitchell, S., 2021. On the dangers of stochastic parrots: Can language models be too big?, in: Proceedings of the ACM Conference on Fairness, Accountability, and Transparency (FAccT), pp. 610–623

    2021

  19. [19]

    Information Management: A Proposal

    Berners-Lee, T., 1989. Information Management: A Proposal. Technical Report. CERN

    1989

  20. [20]

    The semantic web

    Berners-Lee, T., Hendler, J., Lassila, O., 2001. The semantic web. Sci- entific American 284, 34–43. 65

    2001

  21. [21]

    The anatomy of a large-scale hypertextual web search engine, in: Proceedings of the 7th International Conference on World Wide Web (WWW), pp

    Brin, S., Page, L., 1998. The anatomy of a large-scale hypertextual web search engine, in: Proceedings of the 7th International Conference on World Wide Web (WWW), pp. 107–117

    1998

  22. [22]

    Cloudflare blocks AI crawlers by default.https: //blog.cloudflare.com

    Cloudflare, 2025a. Cloudflare blocks AI crawlers by default.https: //blog.cloudflare.com

  23. [23]

    The crawl-to-click gap: Cloudflare data on AI bots, training, and referrals.https://blog.cloudflare.com

    Cloudflare, 2025b. The crawl-to-click gap: Cloudflare data on AI bots, training, and referrals.https://blog.cloudflare.com

  24. [24]

    From Googlebot to GPTBot: Who’s crawling your site in 2025.https://blog.cloudflare.com

    Cloudflare, 2025c. From Googlebot to GPTBot: Who’s crawling your site in 2025.https://blog.cloudflare.com

    2025

  25. [25]

    Google’s AI crawler policy and competitive implica- tions.https://blog.cloudflare.com

    Cloudflare, 2026. Google’s AI crawler policy and competitive implica- tions.https://blog.cloudflare.com

    2026

  26. [26]

    C2PA tech- nical specification.https://c2pa.org/specifications

    Coalition for Content Provenance and Authenticity, 2024. C2PA tech- nical specification.https://c2pa.org/specifications

    2024

  27. [27]

    Mind2web: Towards a generalist agent for the web, in: Advances in Neural Information Processing Systems (NeurIPS)

    Deng, X., Gu, Y., Zheng, B., Chen, S., Stevens, S., Wang, B., Sun, H., Su, Y., 2023. Mind2web: Towards a generalist agent for the web, in: Advances in Neural Information Processing Systems (NeurIPS)

    2023

  28. [28]

    Agent AI: Surveying the horizons of multimodal interaction, in: arXiv preprint arXiv:2401.03568

    Durante, Z., Sarkar, B., Gong, R., Taori, R., Noda, Y., Tang, P., Adeli, E., Lakshmikanth, S.K., Schulman, K., Milstein, A., et al., 2024. Agent AI: Surveying the horizons of multimodal interaction, in: arXiv preprint arXiv:2401.03568

    Pith/arXiv arXiv 2024

  29. [29]

    General data protection regulation (GDPR) – regulation (eu) 2016/679.https://gdpr-info.eu

    European Parliament, 2016. General data protection regulation (GDPR) – regulation (eu) 2016/679.https://gdpr-info.eu

    2016

  30. [30]

    The economics of the online advertising industry

    Evans, D.S., 2008. The economics of the online advertising industry. Review of Network Economics 7, 359–391

    2008

  31. [31]

    Felt, A.P., Barnes, R., King, A., Palmer, C., Bentzel, C., Tabriz, P.,

  32. [32]

    1323–1338

    Measuring HTTPS adoption on the web, in: 26th USENIX Secu- rity Symposium, pp. 1323–1338

  33. [33]

    HTTP/1.1 – RFC 2616

    Fielding, R., Gettys, J., Mogul, J., Nielsen, H., Masinter, L., Leach, P., Berners-Lee, T., 1999. HTTP/1.1 – RFC 2616. Technical Report. IETF. 66

    1999

  34. [34]

    Architectural Styles and the Design of Network- based Software Architectures

    Fielding, R.T., 2000. Architectural Styles and the Design of Network- based Software Architectures. Ph.D. thesis. University of California, Irvine

    2000

  35. [35]

    Agent-to-agent protocol (A2A).https://developers

    Google, 2025. Agent-to-agent protocol (A2A).https://developers. google.com/agent-to-agent

    2025

  36. [36]

    The OAuth 2.0 Authorization Framework – RFC 6749

    Hardt, D., 2012. The OAuth 2.0 Authorization Framework – RFC 6749. Technical Report. IETF

    2012

  37. [37]

    Foundation models and fair use, in: arXiv preprint arXiv:2303.15715

    Henderson, P., Li, X., Jurafsky, D., Hashimoto, T., Lemley, M.A., Liang, P., 2023. Foundation models and fair use, in: arXiv preprint arXiv:2303.15715

    arXiv 2023

  38. [38]

    A proposed standard for LLMs: llms.txt.https: //llmstxt.org

    Howard, J., 2024. A proposed standard for LLMs: llms.txt.https: //llmstxt.org

    2024

  39. [39]

    Infrastructure for AI Agents

    Kapoor, S., et al., 2025. Infrastructure for AI Agents. Technical Report 2501.10114. arXiv

    arXiv 2025

  40. [40]

    Html is the language of the agentic web.https: //twitter.com/karpathy

    Karpathy, A., 2025. Html is the language of the agentic web.https: //twitter.com/karpathy

    2025

  41. [41]

    A watermark for large language models, in: International Conference on Machine Learning (ICML)

    Kirchenbauer, J., Geiping, J., Wen, Y., Katz, J., Miers, I., Goldstein, T., 2023. A watermark for large language models, in: International Conference on Machine Learning (ICML)

    2023

  42. [42]

    A standard for robot exclusion.https://www

    Koster, M., 1994. A standard for robot exclusion.https://www. robotstxt.org/orig.html

    1994

  43. [43]

    Agentic Web: Weaving the Next Web with AI Agents

    Li, Y., et al., 2025. Agentic Web: Weaving the Next Web with AI Agents. Technical Report 2507.21206. arXiv

    arXiv 2025

  44. [44]

    Responsive Web Design

    Marcotte, E., 2011. Responsive Web Design. A Book Apart

    2011

  45. [45]

    Provenance requirements for agentic content pipelines

    Martini, M., et al., 2024. Provenance requirements for agentic content pipelines. arXiv preprint

    2024

  46. [46]

    NLWeb: Natural language interface for the web.https: //github.com/microsoft/NLWeb

    Microsoft, 2025. NLWeb: Natural language interface for the web.https: //github.com/microsoft/NLWeb. 67

    2025

  47. [47]

    Economic models for AI agent content access

    Nguyen, T., et al., 2025. Economic models for AI agent content access. arXiv preprint

    2025

  48. [48]

    User behavior with AI overviews in search

    Pew Research Center, 2025. User behavior with AI overviews in search. https://www.pewresearch.org

    2025

  49. [49]

    HTML 4.01 specification

    Raggett, D., Le Hors, A., Jacobs, I., 1999. HTML 4.01 specification. W3C Recommendation

    1999

  50. [50]

    The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary

    Raymond, E.S., 1999. The Cathedral and the Bazaar: Musings on Linux and Open Source by an Accidental Revolutionary. O’Reilly Media

    1999

  51. [51]

    Agent Communication Protocol (ACP)

    Research, I., 2025. Agent Communication Protocol (ACP). Technical Report. IBM

    2025

  52. [52]

    Zero-click searches in AI mode.https://www.semrush

    Semrush, 2025. Zero-click searches in AI mode.https://www.semrush. com

    2025

  53. [53]

    The curse of recursion: Training on generated data makes models forget, in: Advances in Neural Information Processing Systems (NeurIPS)

    Shumailov, I., Shumaylov, Z., Zhao, Y., Gal, Y., Papernot, N., Ander- son, R., 2023. The curse of recursion: Training on generated data makes models forget, in: Advances in Neural Information Processing Systems (NeurIPS)

    2023

  54. [54]

    Click-through rate analysis: AI overviews impact

    SISTRIX, 2026. Click-through rate analysis: AI overviews impact. https://www.sistrix.com

    2026

  55. [55]

    Agent web: Architecture and challenges.https://www.w3.org/community/agent-web

    W3C Agent Web Community Group, 2025. Agent web: Architecture and challenges.https://www.w3.org/community/agent-web

    2025

  56. [56]

    A survey on large language model based autonomous agents

    Wang, L., Ma, C., Feng, X., Zhang, Z., Yang, H., Zhang, J., Chen, Z., Tang, J., Chen, X., Lin, Y., Zhao, W.X., Wei, Z., Wen, J.R., 2024. A survey on large language model based autonomous agents. Frontiers of Computer Science 18

    2024

  57. [57]

    webMCP: A Client-Side Standard for Agent- Ready Web Design

    webMCP Authors, 2025. webMCP: A Client-Side Standard for Agent- Ready Web Design. Technical Report 2508.09171. arXiv

    arXiv 2025

  58. [58]

    The Attention Merchants: The Epic Scramble to Get Inside Our Heads

    Wu, T., 2016. The Attention Merchants: The Epic Scramble to Get Inside Our Heads. Knopf. 68

    2016

  59. [59]

    Webshop: Towards scalable real-world web interaction with grounded language agents, in: Advances in Neural Information Processing Systems (NeurIPS)

    Yao, S., Chen, H., Yang, J., Narasimhan, K., 2022. Webshop: Towards scalable real-world web interaction with grounded language agents, in: Advances in Neural Information Processing Systems (NeurIPS)

    2022

  60. [60]

    Defending against neural fake news, in: Advances in Neural Information Processing Systems (NeurIPS)

    Zellers, R., Holtzman, A., Rashkin, H., Bisk, Y., Farhadi, A., Roesner, F., Choi, Y., 2019. Defending against neural fake news, in: Advances in Neural Information Processing Systems (NeurIPS)

    2019

  61. [61]

    WebArena: A realis- tic web environment for building autonomous agents, in: International Conference on Learning Representations (ICLR)

    Zhou, S., Xu, F.F., Zhu, H., Zhou, X., Lo, R., Sridhar, A., Cheng, X., Bisk, Y., Fried, D., Alon, U., Neubig, G., 2024. WebArena: A realis- tic web environment for building autonomous agents, in: International Conference on Learning Representations (ICLR)

    2024

  62. [62]

    The Future of the Internet – And How to Stop It

    Zittrain, J., 2008. The Future of the Internet – And How to Stop It. Yale University Press. 69

    2008