LLM Agents Are the Antidote to Walled Gardens
Pith reviewed 2026-05-22 00:36 UTC · model grok-4.3
The pith
LLM-based agents make interoperability between any digital services dramatically cheaper by automatically translating data formats and interacting with human-designed interfaces.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
LLM-based agents can automatically translate between data formats and interact with interfaces designed for humans, which makes interoperability dramatically cheaper and effectively unavoidable. This shift produces universal interoperability: the ability for any two digital services to exchange data seamlessly using AI-mediated adapters. Universal interoperability undermines monopolistic behaviours and promotes data portability, but it can also lead to new security risks, technical debt, and legal frictions.
What carries the argument
LLM-based agents serving as AI-mediated adapters that translate data formats and operate human-designed interfaces to enable universal interoperability.
If this is right
- Any two digital services gain the ability to exchange data seamlessly without custom integrations.
- User lock-in weakens because data portability becomes practical and low-cost.
- Monopolistic practices face pressure as interoperability undercuts barriers to switching services.
- New security risks, accumulated technical debt, and legal frictions emerge as side effects.
- The machine learning community should develop frameworks to address the downsides while supporting the shift.
Where Pith is reading between the lines
- Platforms may respond by designing interfaces that are harder for agents to parse, slowing adoption.
- Widespread use could shift regulatory focus from mandating APIs to overseeing agent-mediated access.
- Smaller services gain easier entry to compete with incumbents through automatic compatibility.
- Over time this could reduce the value of proprietary data silos as exchange becomes default.
Load-bearing premise
LLM agents will reliably and securely interact with arbitrary human-designed interfaces at large scale without high error rates, security failures, or effective resistance from platforms.
What would settle it
A controlled test showing that agents achieve high error rates or trigger platform blocks when attempting to exchange data across major closed services at realistic volumes.
Figures
read the original abstract
While the Internet's core infrastructure was designed to be open and universal, today's application layer is dominated by closed, proprietary platforms. Open and interoperable APIs require significant investment, and market leaders have little incentive to enable data exchange that could erode their user lock-in. We argue that LLM-based agents fundamentally disrupt this status quo. Agents can automatically translate between data formats and interact with interfaces designed for humans: this makes interoperability dramatically cheaper and effectively unavoidable. We name this shift universal interoperability: the ability for any two digital services to exchange data seamlessly using AI-mediated adapters. Universal interoperability undermines monopolistic behaviours and promotes data portability. However, it can also lead to new security risks, technical debt, and legal frictions. Our position is that the ML community should embrace this development while building the appropriate frameworks to mitigate the downsides. By acting now, we can harness AI to restore user freedom and competitive markets without sacrificing security.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper is a position piece claiming that LLM-based agents will disrupt closed proprietary platforms ('walled gardens') by automatically translating data formats and interacting with human-designed interfaces, thereby making interoperability 'dramatically cheaper and effectively unavoidable.' It introduces the term 'universal interoperability' for AI-mediated seamless data exchange between any services, argues this undermines monopolistic lock-in and enables data portability, notes attendant risks (security, technical debt, legal friction), and urges the ML community to develop mitigating frameworks to harness the benefits.
Significance. If the central thesis were substantiated, the work would offer a timely perspective on how agent capabilities could reshape platform economics and user agency in digital services. It correctly identifies a potential shift from explicit API standards to implicit, AI-mediated adapters and flags relevant downside risks. However, the absence of any empirical grounding, failure-mode analysis, or scaling arguments for reliable agent-interface interaction substantially reduces its contribution to the literature on interoperability or agent systems.
major comments (2)
- [Abstract] Abstract: The load-bearing claim that agents 'automatically translate between data formats and interact with interfaces designed for humans: this makes interoperability dramatically cheaper and effectively unavoidable' is asserted without any supporting analysis of error rates, statefulness, UI fragility, or security surface area. Current agent literature routinely documents high failure rates on multi-step or stateful tasks; treating reliable scale as given directly weakens the 'unavoidable' conclusion.
- [Introduction / Definition of universal interoperability] The manuscript introduces 'universal interoperability' as a new capability but provides no concrete mechanism, comparison to existing standards (e.g., OAuth, OpenAPI, or data-portability regulations), or discussion of how platforms could resist or co-opt agent-mediated adapters. This leaves the central disruption argument at the level of assertion rather than technical argument.
minor comments (2)
- [Abstract] The abstract and body use 'universal interoperability' without an explicit definition or scope; a short formalization or set of desiderata would improve clarity.
- [Risks paragraph] The discussion of risks (security, technical debt, legal friction) is high-level; adding one paragraph with concrete examples or references to related work on agent security would strengthen the call for mitigation frameworks.
Simulated Author's Rebuttal
We thank the referee for their constructive and detailed feedback. We value the recognition that the topic is timely and that the paper correctly identifies risks alongside potential benefits. We address each major comment below, noting where revisions will strengthen the manuscript while preserving its nature as a position paper.
read point-by-point responses
-
Referee: [Abstract] Abstract: The load-bearing claim that agents 'automatically translate between data formats and interact with interfaces designed for humans: this makes interoperability dramatically cheaper and effectively unavoidable' is asserted without any supporting analysis of error rates, statefulness, UI fragility, or security surface area. Current agent literature routinely documents high failure rates on multi-step or stateful tasks; treating reliable scale as given directly weakens the 'unavoidable' conclusion.
Authors: We agree that the abstract presents a forward-looking claim without explicit qualification or error analysis. As this is a position paper rather than an empirical study, we focused on the directional argument rather than current performance metrics. To address this point directly, we will revise the abstract to qualify the claim (e.g., noting that reliability is improving but not yet universal) and add a concise discussion of documented agent limitations, including references to failure rates in multi-step tasks from the existing literature. This will make the argument more balanced without altering the core position. revision: partial
-
Referee: [Introduction / Definition of universal interoperability] The manuscript introduces 'universal interoperability' as a new capability but provides no concrete mechanism, comparison to existing standards (e.g., OAuth, OpenAPI, or data-portability regulations), or discussion of how platforms could resist or co-opt agent-mediated adapters. This leaves the central disruption argument at the level of assertion rather than technical argument.
Authors: The manuscript is framed as a high-level position statement to highlight an emerging shift rather than to deliver a systems-level design or exhaustive comparison. We nonetheless recognize the value of grounding the argument more explicitly. In revision we will expand the relevant section to briefly compare universal interoperability to existing mechanisms such as OAuth and OpenAPI, noting their adoption barriers, and to discuss plausible platform countermeasures (rate limiting, legal restrictions on automation, interface changes) together with how agents might adapt, drawing on current practices in web automation. These additions will strengthen the technical framing while keeping the paper concise. revision: yes
Circularity Check
No circularity: forward-looking position argument with no derivations or self-referential reductions
full rationale
The manuscript is a position paper that asserts LLM agents will enable 'universal interoperability' by automatically translating formats and interacting with human-designed interfaces. This claim rests on stated general properties of current LLM agents rather than any equations, fitted parameters, predictions derived from subsets of data, or load-bearing self-citations. No section presents a derivation chain that reduces to its own inputs by construction; the argument is explicitly forward-looking and acknowledges risks without claiming mathematical necessity. The paper therefore contains no instances of the enumerated circularity patterns.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption LLM agents can automatically translate between data formats and interact with interfaces designed for humans at low cost and high reliability.
invented entities (1)
-
universal interoperability
no independent evidence
Forward citations
Cited by 1 Pith paper
-
LLM Harms: A Taxonomy and Discussion
This paper proposes a taxonomy of LLM harms in five categories and suggests mitigation strategies plus a dynamic auditing system for responsible development.
Reference graph
Works this paper leans on
- [1]
-
[2]
R. Agranat and M. S. Gal. Fueling concentration: Ai agents and network effects. Network Law Review, Spring, pages 2016–2023, 2025
work page 2016
-
[3]
A. K. Akanbi and M. Masinde. Semantic interoperability middleware architecture for hetero- geneous environmental data sources. In 2018 IST-Africa Week Conference (IST-Africa), pages Page–1. IEEE, 2018
work page 2018
- [4]
- [5]
-
[6]
L. Bainbridge. Ironies of automation. In Analysis, design and evaluation of man–machine systems, pages 129–135. Elsevier, 1983
work page 1983
-
[7]
J. Bandlamudi, R. Chaudhuri, N. Gantayat, K. Mukherjee, P. Agarwal, R. Sindhgatta, and S. Mehta. A framework for testing and adapting rest apis as llm tools. arXiv preprint arXiv:2504.15546, 2025
-
[8]
P. Béchard and O. M. Ayala. Reducing hallucination in structured outputs via retrieval- augmented generation. arXiv preprint arXiv:2404.08189, 2024
-
[9]
M. Bourreau, J. Krämer, and M. Buiten. Interoperability in digital markets. SSRN Electronic Journal, 2022
work page 2022
-
[10]
N. Carlini, M. Jagielski, C. A. Choquette-Choo, D. Paleka, W. Pearce, H. Anderson, A. Terzis, K. Thomas, and F. Tramèr. Poisoning web-scale training datasets is practical. In 2024 IEEE Symposium on Security and Privacy (SP), pages 407–425. IEEE, 2024
work page 2024
-
[11]
S. Castano, A. Ferrara, and S. Montanelli. Ontology-based interoperability services for semantic collaboration in open networked systems. In Interoperability of Enterprise Software and Applications, pages 135–146. Springer, 2006
work page 2006
-
[12]
T. Chakraborti, V . Isahagian, R. Khalaf, Y . Khazaeni, V . Muthusamy, Y . Rizk, and M. Unuvar. From robotic process automation to intelligent process automation: Emerging trends. arXiv preprint arXiv:2007.13257, 2020
- [13]
-
[14]
S.-R. Cheng. Market power and switching costs: An empirical study of online networking market. U. Cin. L. Rev., 90:122, 2021
work page 2021
- [15]
-
[16]
Business Rules in e-Government Applications
F. Corradini, A. Polzonetti, and O. Riganelli. Business rules in e-government applications. arXiv preprint arXiv:1802.08484, 2018
work page internal anchor Pith review Pith/arXiv arXiv 2018
-
[17]
J. Crémer and G. Biglaiser. Switching costs and network effects in competition policy. In Recent advances in the analysis of competition policy and regulation. Edward Elgar Publishing, 2012. 10
work page 2012
-
[18]
J. Crémer, Y .-A. de Montjoye, and H. Schweitzer. Competition policy, 2019
work page 2019
-
[19]
R. Cyganiak, D. Wood, and M. Lanthaler. Rdf 1.1 concepts and abstract syntax. W3c recommendation, World Wide Web Consortium (W3C), February 2014
work page 2014
-
[20]
A. Delamotte and J. Walter. Akirabot | ai-powered bot bypasses captchas, spams websites at scale. SentinelOne Labs, April 2025. Accessed: 2025-05-21
work page 2025
-
[21]
J.-P. Dubé, G. J. Hitsch, and P. E. Rossi. Do switching costs make markets less competitive? Journal of Marketing research, 46(4):435–445, 2009
work page 2009
-
[22]
B. Engels. Data portability among online platforms. Internet Policy Review, 5(2), 2016
work page 2016
-
[23]
European Commission. Commission decision of 24 march 2004 relating to a proceeding under article 82 of the ec treaty (case comp/c-3/37.792 – microsoft). https://eur-lex. europa.eu/legal-content/EN/TXT/?uri=CELEX:32007D0053, 2004. Official Journal L 32, 6.2.2007, p. 23–28
work page 2004
-
[24]
European Parliament and Council of the European Union. Regulation (eu) 2016/679 of the european parliament and of the council of 27 april 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing directive 95/46/ec (general data protection regulation), 2016. OJ L 119, 4.5....
work page 2016
-
[25]
WASP: Benchmarking Web Agent Security Against Prompt Injection Attacks
I. Evtimov, A. Zharmagambetov, A. Grattafiori, C. Guo, and K. Chaudhuri. Wasp: Benchmark- ing web agent security against prompt injection attacks. arXiv preprint arXiv:2504.18575, 2025
work page internal anchor Pith review Pith/arXiv arXiv 2025
-
[26]
K. R. Fall and W. R. Stevens. Tcp/ip illustrated, volume 1. Addison-Wesley Professional, 2012
work page 2012
-
[27]
J. Farrell and G. Saloner. Converters, compatibility, and the control of interfaces. The journal of industrial economics, pages 9–35, 1992
work page 1992
-
[28]
R. T. Fielding. Architectural styles and the design of network-based software architectures. University of California, Irvine, 2000
work page 2000
-
[29]
C. Fiesler, N. Beard, and B. C. Keegan. No robots, spiders, or scrapers: Legal and ethical regulation of data collection methods in social media terms of service. In Proceedings of the international AAAI conference on web and social media, volume 14, pages 187–196, 2020
work page 2020
-
[30]
R. Frieden. The internet of platforms and walled gardens: Implications for openness and neutrality. Available at SSRN 2754583, 2016
work page 2016
- [31]
- [32]
-
[33]
D. Gourley and B. Totty. HTTP: the definitive guide. " O’Reilly Media, Inc.", 2002
work page 2002
- [34]
-
[35]
D. Hardt. The OAuth 2.0 Authorization Framework. RFC 6749, Oct. 2012
work page 2012
-
[36]
Fast healthcare interoperability resources (fhir), 2025
Health Level Seven International. Fast healthcare interoperability resources (fhir), 2025. Accessed on May 16, 2025
work page 2025
- [37]
-
[38]
J. Howard. The /llms.txt file. https://llmstxt.org/, Sept. 2024. A proposal to standardize the use of an /llms.txt file to provide information to help LLMs use a website at inference time. 11
work page 2024
-
[39]
Ieee standard computer dictionary: A compilation of ieee standard computer glossaries
IEEE. Ieee standard computer dictionary: A compilation of ieee standard computer glossaries. IEEE Std 610, pages 1–217, 1991
work page 1991
-
[40]
IEEE. Information Technology—Portable Operating System Interface (POSIX)—Part 1: System Application Program Interface (API) [C Language]. IEEE, New York, NY , USA, 1996. IEEE Std 1003.1-1996 (ISO/IEC 9945-1:1996)
work page 1996
-
[41]
ISO 20022: Financial Services – Universal Financial Industry Message Scheme
International Organization for Standardization. ISO 20022: Financial Services – Universal Financial Industry Message Scheme. https://www.iso20022.org/about-iso-20022,
-
[42]
ISO 20022 is a multi-part International Standard prepared by ISO Technical Committee TC68 Financial Services. It describes a common platform for the development of messages using a modeling methodology to capture in a syntax-independent way financial business areas, business transactions, and associated message flows
-
[43]
ISO/IEC 2382:2015 Information technology — Vocabulary
International Organization for Standardization and International Electrotechnical Commission. ISO/IEC 2382:2015 Information technology — Vocabulary . ISO, Geneva, Switzerland, 2015
work page 2015
-
[44]
B. John. Challenges and solutions in data integration for heterogeneous systems. 03 2025
work page 2025
- [45]
-
[46]
M. L. Katz and C. Shapiro. Network externalities, competition, and compatibility. The American economic review, 75(3):424–440, 1985
work page 1985
- [47]
-
[48]
G. Klyne and J. J. Carroll. Resource description framework (rdf): Concepts and abstract syntax. W3C Recommendation, 2004
work page 2004
-
[49]
H. Knoche and W. Hasselbring. Continuous api evolution in heterogenous enterprise software systems. In 2021 IEEE 18th International Conference on Software Architecture (ICSA), pages 58–68. IEEE, 2021
work page 2021
-
[50]
Api security perspectives 2025: Ai-enhanced threats and api security
Kong Inc. Api security perspectives 2025: Ai-enhanced threats and api security. Technical report, Kong Inc., December 2024. Accessed: 2025-05-21
work page 2025
-
[51]
R. Lane. Why perplexity’s cynical theft represents everything that could go wrong with ai. Forbes, June 2024. Accessed: 2025-05-21
work page 2024
-
[52]
R. Lehmann. Towards interoperability of apis-an llm-based approach. In Proceedings of the 25th International Middleware Conference: Demos, Posters and Doctoral Symposium, pages 29–30, 2024
work page 2024
-
[53]
C. Lemmer-Webber, J. Tallon, E. Shepherd, A. Guy, and E. Prodromou. Activitypub. https://www.w3.org/TR/2018/REC-activitypub-20180123/, Jan. 2018. W3C Rec- ommendation
work page 2018
-
[54]
A. Lercher, J. Glock, C. Macho, and M. Pinzger. Microservice api evolution in practice: a study on strategies and challenges. Journal of Systems and Software, 215:112110, 2024
work page 2024
- [55]
- [56]
-
[57]
D. Mehrotra and T. Marchman. Perplexity is a bullshit machine. WIRED, June 2024
work page 2024
-
[58]
S. M. Mirtaheri, M. E. Dinçktürk, S. Hooshmand, G. V . Bochmann, G.-V . Jourdan, and I. V . Onut. A brief history of web crawlers. arXiv preprint arXiv:1405.0749, 2014. 12
work page internal anchor Pith review Pith/arXiv arXiv 2014
-
[59]
A. Mohammadjafari, A. S. Maida, and R. Gottumukkala. From natural language to sql: Review of llm-based text-to-sql systems. arXiv preprint arXiv:2410.01066, 2024
- [60]
-
[61]
M. Nagarajan, K. Verma, A. P. Sheth, J. Miller, and J. Lathem. Semantic interoperability of web services-challenges and experiences. In 2006 IEEE International Conference on Web Services (ICWS’06), pages 373–382. IEEE, 2006
work page 2006
-
[62]
Competition in the mobile application ecosystem
National Telecommunications and Information Administration. Competition in the mobile application ecosystem. Technical report, U.S. Department of Commerce, February 2023. Accessed: 2025-05-20
work page 2023
-
[63]
F. Neuhaus and J. Hastings. Ontology development is consensus creation, not (merely) representation. Applied Ontology, 17(4):495–513, 2022
work page 2022
-
[64]
Data portability, interoperability and digital platform competition: Oecd background paper
OECD. Data portability, interoperability and digital platform competition: Oecd background paper. Interoperability and Digital Platform Competition: OECD Background Paper (June 8, 2021), 2021
work page 2021
-
[65]
Open Banking Working Group. The open bank- ing standard. https://psr.org.uk/media/004p1ws5/ background-document-no-2-the-open-banking-standard-full-report.pdf ,
-
[66]
Commissioned by HM Treasury
-
[67]
Openapi specification version 3.1.1, October 2024
OpenAPI Initiative. Openapi specification version 3.1.1, October 2024. Published by the OpenAPI Initiative, a project of the Linux Foundation
work page 2024
-
[68]
E. Parliament. Regulation (eu) 2022/1925 of the european parliament and of the council of 14 september 2022 on contestable and fair markets in the digital sector and amending directives (eu) 2019/1937 and (eu) 2020/1828 (digital markets act). https://eur-lex.europa.eu/ eli/reg/2022/1925/oj, October 2022. OJ L 265, 12.10.2022, pp. 1–66
work page 2022
-
[69]
S. G. Patil, T. Zhang, X. Wang, and J. E. Gonzalez. Gorilla: Large language model connected with massive apis. Advances in Neural Information Processing Systems, 37:126544–126565, 2024
work page 2024
-
[70]
A. Plesner, T. V ontobel, and R. Wattenhofer. Breaking recaptchav2. In2024 IEEE 48th Annual Computers, Software, and Applications Conference (COMPSAC), pages 1047–1056. IEEE, 2024
work page 2024
-
[71]
E. Prud’hommeaux and A. Seaborne. Sparql query language for rdf. W3C Recommendation, 2008
work page 2008
- [72]
- [73]
-
[74]
Y . Ruan, H. Dong, A. Wang, S. Pitis, Y . Zhou, J. Ba, Y . Dubois, C. J. Maddison, and T. Hashimoto. Identifying the risks of lm agents with an lm-emulated sandbox. arXiv preprint arXiv:2309.15817, 2023
work page internal anchor Pith review Pith/arXiv arXiv 2023
-
[75]
J. Ryan. Will ai finally destroy the advertising-based business model of the internet? Journal of Business and Artificial Intelligence, 1(1):n/a, May 2024. Commentary
work page 2024
-
[76]
A. Saha, L. Mandal, B. Ganesan, S. Ghosh, R. Sindhgatta, C. Eberhardt, D. Debrunner, and S. Mehta. Sequential api function calling using graphql schema. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages 19452–19458, 2024. 13
work page 2024
-
[77]
P. Saint-Andre. Extensible messaging and presence protocol (xmpp): Core. Technical report, 2011
work page 2011
-
[78]
N. Sakimura, J. Bradley, M. Jones, B. de Medeiros, and C. Mortimore. Openid connect core 1.0. OpenID Foundation, 2014
work page 2014
- [79]
-
[80]
A. Searles, R. T. Prapty, and G. Tsudik. Dazed & confused: A large-scale real-world user study of recaptchav2. arXiv preprint arXiv:2311.10911, 2023
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.