◆ Semantic Web Catalogue

Problem thread

Agents acting for us

Will software agents read the Web's data and act on people's behalf, and what do they need from it?

issue
spans all issues
status
open
steps
7, 1998–2026
threads
all problem threads

The best-known promise of the Semantic Web was software agents that would read machine-readable meaning on the Web and run errands for people. By 2007 one of the vision's authors was asking where those agents were. Developers built web APIs instead, from REST to GraphQL. Language-model agents have brought the promise back, connecting to tools and data through protocols such as MCP rather than through shared ontologies, and a W3C group is again working on Web-native descriptions of what agents can do.

Lineage

  1. Sep 1998
    precedent writing ★ seminal
    Semantic Web Road Map

    A road map toward a Web in which machine reasoning will be ubiquitous and devastatingly powerful, built on languages for expressing information in machine-processable form. [src]

  2. 2000
    precedent writing ★ seminal
    Architectural Styles and the Design of Network-based Software Architectures

    Defines REST, the Web's architectural style: resources named by URIs and manipulated through representations over a uniform interface. Most web APIs since claim descent from it. [src]

  3. May 2001
    proposes writing ★ seminal
    The Semantic Web

    Software agents that act on machine-readable meaning published on the Web, finding and booking services for the people they work for. [src]

  4. May 2007
    measures writing
    Where Are All the Intelligent Agents?

    Six years after co-writing the 2001 article, Hendler asks in an editorial where all the intelligent agents are. [src]

  5. 14 Sep 2015
    workaround standard
    GraphQL

    A query language for web APIs in which each client asks for exactly the JSON it needs from one application's own schema, with no shared vocabulary between sites. [src]

  6. 25 Nov 2024
    proposes standard
    Model Context Protocol

    An open standard for connecting AI applications to data sources, tools and workflows, so agents can access information and act on a user's behalf. [src]

  7. 20 Oct 2026
    proposes writing
    W3C and the State of the Web in the AI Era

    The W3C's strategy paper finds that agents prefer text and structured data to HTML, that MCP is unlikely ever to offer endpoints at the scale of websites, and that JSON-LD, RDF, DCAT and ODRL become more valuable as inputs and guardrails for agents, with Verifiable Credentials and DIDs to identify agents and whom they act for. [src]

Recent signals

Dated events that show the problem moving, newest last.

  1. 9 Dec 2025

    Anthropic gives the Model Context Protocol to the new Agentic AI Foundation under the Linux Foundation, co-founded with Block and OpenAI; it counts more than 10,000 active public MCP servers. Model Context Protocol[src]

  2. 17 Aug 2026

    Google's Agent2Agent protocol, for agents talking to one another, moves into the same foundation, which has grown to more than 250 members.[src]

Open now

Under each question, the answers people have put forward or built. The catalogue gives no odds on them.

  • Should agents rely on shared vocabularies and published data, or on each service's own tools and APIs? [src1][src2]
    MCP connects agents to tools and data source by source; the W3C AI Agent Protocol Community Group is working on Web-native metadata describing agent capabilities to support automated reasoning.
    • Agents read shared vocabularies and published structured data. W3C and the State of the Web in the AI Era
      The W3C's 2026 paper argues that agents prefer text and structured data to HTML, so JSON-LD, RDF, DCAT and ODRL become more valuable as inputs.[src]
    • Each service offers its own tools to agents through a common protocol. Model Context Protocol
      MCP standardizes how an AI application calls a service's tools and reads its data; the vocabulary of each tool stays the service's own.[src]
  • How should an agent acting for someone decide which tools and data to trust? [src]
    The MCP documentation includes security guidance for servers and clients.
    • Agents and the people they act for carry verifiable identities. Verifiable Credentials Data Model
      The same W3C paper proposes Verifiable Credentials and DIDs to identify agents and whom they act for.[src]

Forces

Pressures that push this problem one way or another, as their sources argue them. All forces

How to contribute