Agentic AI Knowledge Domains — Category Map/Taxonomy

DRAFT · June 2026   W3C AI Knowledge Representation Community Group (AI KR CG)

Invitation to contribute. You are invited to review these categories, propose changes, and discuss them on the public mailing list of the AI KR CG: public-aikr@w3.org — archives at lists.w3.org/Archives/Public/public-aikr/. Anyone may read or write to this list.
Provisional — to be confirmed and maintained. The agentic-AI environment is changing rapidly. Every term, classification, and link in this map is provisional and must be verified and kept current to remain useful. This is a living draft, not a settled reference. Entries whose source is not yet confirmed are marked to confirm.

1. How these categories were identified

The categories summarise terms, standards, and standards bodies distilled from resources describing the agentic-AI landscape (see also the Vocabulary and the Standards landscape), and are open for review; they were derived by reconciling the group's six-dimension agent ontology (what an agent is) with the five-section vocabulary of the surrounding ecosystem, plus an eighth category for applied, feed-derived material.

2. How to use this map

This is the top-level knowledge domain for the repository's agentic-AI content. The two contributable files — the Vocabulary and the Standards inventory — are organised under these headings. All other repository files are references. Each term below links to its source: an external live resource where one exists, or inward to the group's own vocabulary where the AI KR CG is itself the source.

3. The eight categories

#CategoryScopeExample terms (each linked to source)
1Agent types & autonomyKinds of agent and their degree of independence. ReactiveAgent, DeliberativeAgent, LearningAgent, AutonomousGoalSeekingAgent, AssistantAgent, AnalyticalAgent, CreativeAgent, TaskAgent
2Cognition & capabilityWhat an agent can do internally, plus its form and reach. Perception, Planning, Reasoning, Learning, Memory, embodiment, operational scope
3Interaction, communication & identityHow agents coordinate, message, and are identified. A2A (Agent2Agent), MCP (Model Context Protocol), ACP (Agent Communication Protocol)[a], ANP (Agent Network Protocol), OASF (Open Agent Schema Framework), NANDA[b], DID (Decentralized Identifiers), Verifiable Credentials, human-in-the-loop, delegation
4Commerce & settlementHow agents transact and how value clears. x402, AP2 (Agent Payments Protocol), ACP (Agentic Commerce Protocol)[a], ERC-8004 (Trustless Agents)[c], agentic clearing[d], settlement determination, stablecoin
5Oversight, safety & interpretabilityHow agent action is reviewed, constrained, and understood. OWASP Agentic Security Initiative, OWASP Top 10 for Agentic Applications, CSA MAESTRO[e], graduated oversight (GAIE)[f], agent-as-judge, interpretability
6Regulation & governanceThe legal and policy frameworks that bind agents. EU AI Act / GPAI[g], DORA[h], GENIUS Act[i], NIST CAISI AI Agent Standards Initiative[j], IMDA Model AI Governance Framework for Agentic AI[k], Five Eyes joint guidance[l]
7Agentic standards & standards bodiesThe agentic-AI standards and the organisations that maintain them (general semantic-web standards excluded). W3C AI KR CG, W3C AI Agent Protocol CG, W3C AI Agent Memory Interoperability CG[m], Linux Foundation Agentic AI Foundation[n], AGNTCY collective, C2PA (content provenance), CEN-CENELEC JTC 21
8Applied domainsFeed-derived material where agents are applied to a field. quantitative forecasting, reflexivity, Sharpe ratio, biotech, applied coding

4. Notes & disambiguations

  1. [a] "ACP" is overloaded — three distinct, current efforts. Agent Communication Protocol (IBM Research / AGNTCY collective, REST-native), Agent Connect Protocol (Cisco / AGNTCY), and Agentic Commerce Protocol (Stripe + OpenAI). Never use bare "ACP"; always qualify.
  2. [b] NANDA — MIT Media Lab initiative for an "Internet of AI Agents"; DID-based agent discovery/registry (nanda.media.mit.edu). The acronym is expanded two ways in NANDA's own materials — "Networked AI Agents in a Decentralized Architecture" and "Networked Agents and Decentralized AI".
  3. [c] ERC-8004 — "Trustless Agents" Ethereum proposal (EIP-8004) for on-chain agent identity/reputation; a Draft EIP, with reference deployments live on Ethereum mainnet since January 2026.
  4. [d] RAILS — source framework for the clearing/settlement terms; acronym expansion unconfirmed pending source-paper verification.
  5. [e] MAESTRO — Multi-Agent Environment, Security, Threat, Risk & Outcome; a seven-layer agentic threat model (Cloud Security Alliance).
  6. [f] GAIE — source framework for graduated oversight; acronym expansion unconfirmed pending source-paper verification.
  7. [g] GPAI here = General-Purpose AI (EU AI Act, Regulation (EU) 2024/1689); not the Global Partnership on AI.
  8. [h] DORA here = Digital Operational Resilience Act (Regulation (EU) 2022/2554, financial sector); not DevOps Research & Assessment.
  9. [i] GENIUS Act — US stablecoin statute — Public Law 119-27 (S.1582), enacted 18 July 2025.
  10. [j] NIST CAISI — Center for AI Standards and Innovation; AI Agent Standards Initiative launched 17 Feb 2026.
  11. [k] IMDA — Singapore Infocomm Media Development Authority, Model AI Governance Framework for Agentic AI (updated v1.5, 20 May 2026).
  12. [l] Five Eyes joint guidance. The "Five Eyes" are the five allied nations — United States, United Kingdom, Canada, Australia and New Zealand — whose security agencies issue joint cyber-security advisories. Around 1 May 2026 their agencies (six in total, the US fielding both CISA and NSA) issued joint guidance on securely deploying agentic AI, now widely treated as a de-facto operational baseline where no binding regulation yet exists. The primary joint guidance, "Careful Adoption of Agentic AI Services", was published by CISA, the NSA and partner agencies (Australia, Canada, New Zealand, UK) on 1 May 2026 (linked).
  13. [m] W3C AI Agent Memory Interoperability CG — proposed 18 May 2026.
  14. [n] Linux Foundation Agentic AI Foundation (AAIF) — now stewards both MCP and A2A (aaif.io).

Files structured by this map: the Vocabulary and the Standards inventory. All other repository files are references. Status: DRAFT — provisional and subject to ongoing maintenance. License: CC BY 4.0.