WORKING DRAFT   This is a non-normative summary and index of the work of the W3C AI Knowledge Representation (AI KR) Community Group, and a map to the documents held in its repository. The AI KR CG applies canonical knowledge-representation techniques — formal semantics, logic, and controlled natural language — to track, map and evaluate the fragmented and rapidly evolving landscape of agentic AI on the Web and its neighbouring standards. The documents in the repository are working technical notes: temporary flags that record how web agents and their standards are evolving, rather than finished specifications. This note sets out the organising framework into which those notes — and future contributions — fit, and points to an interactive map of how they relate to one another and to neighbouring W3C and IETF work.

This is a draft Community Group document. It is non-normative and editorial in nature; it makes no conformance claims and uses no RFC 2119 requirement language. The technical notes it indexes are snapshots of an evolving landscape — part of an ongoing process of representation and observation, not a finished product. Comments, corrections and contributions are welcome through the Community Group and its repository.

Introduction — why the group exists

Agentic AI is triggering a paradigm shift in computation, delivering unprecedented capabilities together with unprecedented risks — amplified by the immediacy and pervasiveness of the World Wide Web as the substrate on which web agents are developed, tested and deployed. Relevant standards are being produced at great cost across disparate venues — W3C Community and Working Groups, the IETF, and industry consortia — without manifesting coherence between them. The AI KR CG exists to make that landscape legible.

What problem does the group address?

The agentic-AI landscape is fragmented and asymmetric across protocols, vocabularies, architectures and governance regimes. Through canonical KR techniques the CG builds shared, explicit knowledge representations — ontic maps — of the facets of agentic AI, in order to support alignment, consistency and model integrity across these disparate efforts.

What risk does it address?

Fragmentation produces misrepresentation: partial, inconsistent models of what agents are and do, held by systems and people alike. Left implicit, misrepresentation, misalignment and systemic deviation accumulate into systemic risk. By rendering the landscape explicit and inspectable, the CG supports the explainability required for trust, security and reliability.

What gap in current Web technology does this address?

There is no shared, system-level knowledge representation for the agentic-AI ecosystem. The CG contributes one by tracking and exposing the ecosystem's evolution through KR artefacts — concept maps, graphs, vocabularies, ontologies and semantic models — paying particular attention to where representations diverge. The work is devised to support the cognition of generalists and a shared understanding of a vast and complex ecosystem for which there is, as yet, no map.

How to read this note

This is an index and a scaffold, not a specification. Each document in the repository is a focused snapshot taken at the point where the AI KR perspective met a specific neighbouring effort. The two sections that follow give the constructs every note plugs into: the AIAO classification vocabulary (what facet of agentic AI a note concerns) and the Agent Interoperability Framework (which layer of the representation-to-trust progression it serves). The repository map then slots the current documents into that scaffold, and the slot template shows how a new note joins.

Organising framework

AIAO — the agent-classification vocabulary

AIAO synthesises recurring agent taxonomies from the research literature into a single multi-dimensional vocabulary. Its stated design principle is synthesis over invention: existing classifications are harmonised rather than replaced. In its current revision it comprises eight dimensions:

  1. Autonomy Level — Reactive · Deliberative · Learning · AutonomousGoalSeeking
  2. Functional Role — Assistant · Analytical · Creative · Task
  3. Cognitive / Operational Capability — Perception · Planning · Reasoning · Learning · Memory
  4. Interaction Style — HumanInTheLoop · MultiAgentSystem · Collaborative
  5. Embodiment — PureSoftware · Robotic · Hybrid
  6. Operational Scope — SingleTask · MultiTask · OpenEnded
  7. Interface Modality — Text · Voice · Visual · Multimodal
  8. Interaction Paradigm — Transactional · Relational · HybridParadigm

Dimensions 7 and 8 (highlighted) are recent additions contributed by the CG beyond the synthesised literature.

New ontic-map categories surfaced by the CG

Working through concrete interactions (for example the agent-interface and proof binding exchanges), the CG has identified two categories that have no adequate home in prior upper-level schemes and are proposed as genuine additions to the ontic map:

Agent Interoperability Framework (summary)

The framework arranges the work as a progression of layers, from making things explicit to making them trustworthy. Each repository note serves one or more layers:

  1. Representation — explicit, shared representations of agent facets (the KR base).
  2. Alignment — reconciling divergent vocabularies and ontologies so models agree.
  3. Interoperability — enabling agents, protocols and services to operate together on a common model.
  4. Trust — provenance, verification and attestation that make interoperation accountable.

A fuller treatment of the framework is in preparation for separate publication.

Repository map

The table below slots the documents currently in the repository into the scaffold. The ontic category column names the relevant AIAO dimension or ontic-map category; the layer column gives the framework layer the note principally serves; and bridges to names the neighbouring standard it engages.

DocumentOntic categoryLayerBridges to
agonalign — Agent Ontology Alignment (IDL ↔ HMAS) Agent ontologyAlignmentHMAS / Web of Things
webMCP — WebMCP Technical Notes Tool / context invocation (Functional Role · Cognitive)RepresentationModel Context Protocol
conduit — Missing Layer Zero Capability / commerce (Interaction Paradigm: Transactional)Interoperability + TrustIETF VCAP / AIVS / ATEP · Google AP2
voiceAI — Configurable Voice Interaction Interface Modality: VoiceInteroperabilityW3C Web Speech API
DFMC — Digital Forensics Model Card Verification / Trust Layer (provenance)TrustModel Cards
Solid — How to Use Solid Data sovereigntyTrust (data)Solid / decentralised data

Slot template — how a new note joins

A contributor does not need to understand the whole repository to add to it. A new note declares five fields, and it drops into the map:

FieldMeaning
TitleThe note's subject.
Ontic categoryWhich AIAO dimension or ontic-map category it represents.
LayerRepresentation · Alignment · Interoperability · Trust.
Bridges toThe neighbouring W3C / IETF / external standard it engages.
StatusDraft snapshot · under discussion · stable.

Neighbouring standards landscape (short)

The AI KR CG work sits among many related efforts across standards bodies. The table below is a short orientation; a wider, working-draft inventory — with each effort's core elements and a sketch of its AIAO / KR alignment — is maintained separately at Inventory of Agentic-AI Standards Activity.

Body / groupEffortAIAO / KR pointer
W3C — Autonomous Agents on the Web CGHMAS / hypermedia agentsEmbodiment · Alignment
W3C — Agent Protocol CGAgent protocolsInteraction Style · Interoperability
W3C — AIVS CG (Sanctuary)Agentic integrity verificationVerification / Trust Layer
W3C — Web ML WGWebMCP (proposed) / WebNNTool-context invocation · Representation
W3C — Voice / Smart Voice AgentsVoice agent standardsInterface Modality: Voice
W3C — VC / DID WGsCredentials & identityVerification / Trust Layer
IETF — draft-stone-vcap / -atep / -aivsCapability, reputation, integrityInteroperability + Trust · cite AI KR CG TN
ISO/IEC JTC 1 SC 42AI risk / managementGovernance / systemic-risk facet
A2A (Google→LF) · MCP (Anthropic→LF)Agent & tool interoperability — Linux Foundation, maintainer-gatedInteraction Paradigm · Representation → Interoperability

Visual map

An interactive map of how the documents relate to one another and to neighbouring W3C and IETF work is published at https://w3c-cg.github.io/aikr/map/. It is intended to orient new readers and contributors.

Contributing

The most direct way to understand the meaning and impact of this work is to participate in the Community Group. Contributions to the repository are welcome: open an issue or a pull request, populate the slot template for any new note, and the change will be triaged by the maintainer. Discussion takes place on the public-aikr mailing list.

References (informative)

Repository documents: agonalign, webMCP, conduit, voiceAI, DFMC, Solid.

Neighbouring W3C activity: Autonomous Agents on the Web CG, Agent Protocol CG, Web Machine Learning WG (WebMCP), Smart Voice Agents Workshop, Solid CG.

Neighbouring IETF activity: draft-stone-vcap, draft-stone-atep, draft-stone-aivs.