9:30 - 9:40: Agenda bashing
9:40 - 9:50: Chairs' welcome, opening and panellists' introductions
9:50 - 10:10 : Panellist Statements on the topic: "How does standardisation look in an LLM-enabled future?" (7 mins each)
10 - 11 : Discussion: lead by the chairs and for the panellists to discuss with each other.
11- 11:30 : Q&A
Bios
Jie Bian is a Ph.D. candidate at the University of Oslo. Her research project lies at the intersection of Internet protocol design and the machine reading of standardization and discussion documents.
Jaime Jiménez is a Master Researcher at Ericsson with over 15 years of experience in IoT and Web technologies. He co-chairs the IETF CoRE Working Group, contributing to the Constrained Application Protocol (CoAP) and the device management protocol LwM2M. Currently, he is developing Autonomous Generative AI Agents, enhancing their reasoning and integration with Web/IoT APIs and real-time data.
Joseph Potvin is the co-founding Executive Director of the not-for-profit Xalgorithms Foundation (with Bill Olders of DataKinetics), where he leads design research on the "Data With Direction Specification (DWDS)". This ‘rules-as-data’ method enables anyone, on any platform, to author, publish, discover, fetch, scrutinize, prioritize and, optionally, to automate normative data across a network. At IETF105 HotRFC in 2019 he asked if this might potentially enable 'an Internet of Rules'. The specification is explained and detailed in his 2023 doctoral dissertation, with Nhamo Mtetwa as external examiner, and a working reference implementation created by Don Kelly and Huda Hussain. This community effort won Xalgorithms the Fintech/Regtech "Finnovator for 2024" Global Award from Central Banking Publications.
Jean F. Quéralt is the Founder & CEO of The IO Foundation, a nonprofit advocating for Data-Centric Digital Rights dedicated to establishing a more solid approach to Digital Rights from a technical standards perspective.
Abstracts/statements
*Jie Bian
RFCs don’t usually contain much text that explains the reasons behind the design decisions that were taken. Yet, making these rationales visible is vital for several reasons: it supports the newcomers to the IETF, fosters transparency and trust, and enables more informed participation. When the underlying reasonings are clearly articulated, newcomers gain insight into trade-offs and historical context, while implementers can better align their work with the intent of the specification rather than relying solely on prescriptive rules. Beyond immediate benefits, preserving and exposing design rationales is crucial for the evolution of next-generation protocols. Without access to the considerations that shaped earlier standards, future designers risk repeating past mistakes or overlooking subtle compromises that influenced prior decisions.
Building on the above work, we also explored leveraging LLMs for auto-editing RFCs, enabling models to refine draft text based on feedback from WG members. This capability has the potential to significantly accelerate iterative development of Internet-Drafts and streamline IETF workflows.
*Jaime Jiménez
This presentation demonstrates how AI agents can assist with IETF specification making process, based on practical experience. It covers where AI can immediately help with examples. It also discusses practical adoption steps and challenges.
*Joseph Potvin
LLM-Assisted Transformation of IETF Standards into Machine-Processable Rules-as-Data
Developers are challenged to maintain awareness and practical applicability of requirements embedded in the prose of standards, such as more than 9,000 RFCs/BCPs. The Data With Direction Specification (DWDS) is a framework for transforming normative rules into consistent, canonical 'Rules-as-Data' optimized for machine processing, while also being human-auditable. During unit testing, developers could get specific "in effect", "applicable" and "invoked" requirements in standards dynamically discovered, fetched, delivered inside existing developer toolchains, for conformance support and validation. The free/libre/open source reference implementation involves: Rule-as-Data (canonical data structures), RuleMaker (authoring IDE), RuleReserve (IPFS-based distributed storage), and RuleTaker (edge-resident embedded discovery component). Rules-as-Data will be illustrated with a BGP-4 propagation rule example from RFC 4271.
We use LLMs as linguistic tools (not oracles) with human-in-the-loop iteration to: (1) distill RFC prose into discrete declarative statements of Input Conditions and Output Assertions; (2) structure each with a six-element finite grammar; (3) generate a concise interpretive explanation; and (4) cycle human-validated results back into LLMs to iteratively improve performance with (1), (2) and (3) for emergent scalability. Current R&D involves integration of the Model Context Protocol.