The Agentcities initiative did not unfold in isolation. It ran during precisely the years when a parallel and equally ambitious vision was taking shape on the open web — the Semantic Web — and the two communities found themselves circling the same fundamental problem from different directions. Both were trying to make machines capable of understanding meaning well enough to act on it autonomously, and as the initiative matured, the question of how agent technology and Semantic Web technology related to one another moved from the margins to the center of its agenda.
Tim Berners-Lee's Semantic Web Vision
The vision was articulated most famously in a 2001 Scientific American article co-authored by Tim Berners-Lee, the inventor of the World Wide Web. Its premise was that the web of the time consisted of documents designed for humans to read, with markup that described how to display content but said nothing about what the content meant. The Semantic Web proposed to layer machine-interpretable meaning over that document web, transforming it into a web of data in which the relationships between pieces of information were made explicit and formal. The promise was that software could then traverse this web of data, draw inferences, and accomplish tasks on a user's behalf — a vision that mapped naturally onto the agent research community's long-standing ambitions.
W3C Semantic Web Activity
To realize that vision, the World Wide Web Consortium developed a stack of formal standards in parallel with the Agentcities work. The Resource Description Framework (RDF) provided the base data model, representing all knowledge as simple subject-predicate-object triples that could be combined into arbitrarily large graphs. The Web Ontology Language (OWL) built on RDF to express formal ontologies with a rigorous description-logic semantics, allowing machines not merely to store assertions but to reason over them and derive new facts. And SPARQL supplied a standardized query language for interrogating RDF graphs. Together these specifications gave the Semantic Web community a coherent technical foundation, and their development closely tracked the period in which Agentcities was operating.
FIPA Agents and Semantic Technologies
It is easy to forget that FIPA's own ontology model predated the W3C Semantic Web stack. The agent community had already recognized that for two agents to communicate meaningfully they needed shared, formally specified vocabularies, and FIPA's content languages — including SL (Semantic Language), KIF (Knowledge Interchange Format), and Prolog — served much the same purpose that RDF and OWL would later serve, namely the formal representation of knowledge inside agent messages. The relationship between these two parallel efforts was the explicit concern of the Ontology Working Group, which examined how FIPA ontologies might be aligned with the emerging W3C standards rather than competing with them. Building OWL-based ontologies directly into FIPA agent systems became an active research direction, reflecting a growing consensus that the two stacks were complementary expressions of a single underlying goal.
Semantic Web Services
The convergence sharpened around the idea of semantic web services — the marriage of the web services infrastructure that industry was rapidly adopting (WSDL for interface description, SOAP for messaging, and BPEL4WS for orchestration) with the semantic descriptions that ontology languages made possible. Two major efforts pursued this synthesis. OWL-S (OWL for Services), a largely US-led initiative, described services in terms of their preconditions, effects, and internal processes, so that a service's behavior could be reasoned about formally rather than merely invoked blindly. WSMO (the Web Service Modeling Ontology) represented a European approach to the same problem, developed by groups that overlapped substantially with the Agentcities community. The shared aspiration was striking in its ambition: agents that could automatically discover the services they needed, compose several of them into a workflow, and invoke them — all by reasoning over their semantic descriptions, without a human integrating them by hand.
The openNet Vision and NY 2004 Convergence
This widening scope was made explicit at the openNet Forum meeting held in New York in 2004, which directly addressed the convergence of agent technology with the Semantic Web, with Web Services, and even with GRID computing. The very name openNet signalled the expansion of the mandate: what had begun as a project about FIPA agent interoperability was reframing itself as an effort to build a broad open infrastructure spanning multiple technology families. That reframing reflected a hard-won recognition within the community that these technologies were converging rather than competing, and that the future lay in a unified open network rather than in any single one of them prevailing over the others.
Legacy in Contemporary AI
Neither Agentcities nor the Semantic Web ever fully solved the problem at the heart of both efforts — enabling autonomous software systems to share, in a machine-understandable way, a description of their own capabilities and limitations so that other systems could reliably make use of them. It is a striking continuity that the large language model agents of today confront exactly the same fundamental questions about shared ontologies and semantic interoperability that Agentcities wrestled with two decades earlier. The vocabulary has shifted — we now speak of tool definitions, function calling, and agent protocols rather than Directory Facilitators and FIPA-ACL — but the underlying research agenda is remarkably unchanged. Anyone designing how a modern AI agent should advertise its capabilities, negotiate with another agent, or compose external services is working within a problem space that the Agentcities researchers mapped out in considerable detail. For the formal standards that grew out of this era and continue to evolve, the W3C Semantic Web activity remains the canonical reference.