Two journeys · 60 years · one recurring pattern
The future was waiting
for an interface
A technology can be fundamentally sound and still go unappreciated for decades, until a complementary innovation makes it usable by everyone. It happened to hypertext. It is happening right now to the Semantic Web Project.
Chapter 0 · The idea in one picture
Hype usually comes first. Here, value did.
In a classic hype cycle, attention races ahead of value. In an inverted one, value is there for decades while attention stays flat.
Value vs. attention, across one journey
Conceptual illustration, not measured data. The shaded gap is the time a working idea spends waiting for its interface.
Chapter 1 · 1965–1994 · The first inverted hype cycle
The Web didn’t invent hypertext. It liberated it.
Nearly three decades separate the naming of hypertext from its arrival as a mass-market experience. The ideas worked the whole time.
Two journeys, drawn to the same 33-year scale
Tap any point, or use ← → keys. Green marks the arrival of an accessible client.The takeaway
Second-rail dates after 1999 are added context: the 2001 Scientific American article, Linked Data principles (2006), SPARQL’s W3C Recommendation (2008), and mainstream LLM chat (late 2022).
The mouse was an input device. Hypertext was the information architecture.
On December 9, 1968, Douglas Engelbart gave what became known as the Mother of All Demos: interactive editing, collaborative computing, videoconferencing, hypertext navigation in NLS, and the mouse. Popular memory kept the mouse. The deeper message, a computing environment where information is organized, connected, and navigated, took decades to land.
The 1968 demo: what people remember vs. what mattered
The architecture
Information organized, connected, and navigated interactively. Its full potential took decades to realize.
An interface can attract attention while the architecture underneath waits for its moment.
Berners-Lee’s breakthrough wasn’t hypertext. It was removing the platform boundary.
At CERN in 1989–1991, Tim Berners-Lee built a loosely coupled architecture that let hypertext documents be identified, retrieved, and connected across heterogeneous computers. Nobody needed the same application, operating system, or storage. He also created the first browser/editor and server, but that graphical client wasn’t yet the widely distributed experience that would popularize the Web.
Three standards, one architecture
Tap each pieceHypertext was no longer confined to a particular application. It became a capability of the Web.
Then, in 1993–1994, people could simply click.
Mosaic (1993) and Netscape Navigator (1994), both led by Marc Andreessen (Mosaic at NCSA, then Netscape as co-founder), made the Web accessible through familiar graphical interfaces on widely used platforms. The architecture and the open standards already existed. The missing piece was an accessible client that connected them to everyday life.
Try the 1993 experience
One click. That’s the whole point.Mosaic and Netscape didn’t invent hypertext. They turned its utility into a mass-market experience.
The Web’s success wasn’t simply a victory for HTML. It was a victory for loose coupling, standardized resource identification, open protocols, and accessible interfaces.
And history is repeating itself.
Chapter 2 · 1999 → today · The second inverted hype cycle
From hyperlinked documents to hyperlinked data
Hyperlinks don’t just identify documents. They identify the people, organizations, products, and concepts inside them, and software computes over their relationships.
That is the direction pursued by the Semantic Web Project. RDF is not an extension of HTTP. It extends the Web’s resource-identification and linking architecture beyond HTML documents to machine-computable descriptions of entities and relationships.
HTTP provides resource access. HTML provides document presentation. RDF provides a standard model for machine-computable entity relationships.
Same facts, two Webs
Tim Berners-Lee founded the W3C, which publishes RDF, a creative work whose category is defined by the schema.org vocabulary.
HTML links connect documents. A person can read the relationships; software sees words and links between pages.
@prefix dbr: <http://dbpedia.org/resource/> . @prefix schema: <http://schema.org/> . @prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> . dbr:World_Wide_Web_Consortium schema:founder dbr:Tim_Berners-Lee . dbr:Resource_Description_Framework schema:publisher dbr:World_Wide_Web_Consortium ; a schema:CreativeWork . schema:CreativeWork rdfs:isDefinedBy <http://schema.org/> . # the ontology
IRIs identify entities and relationships, so these statements can be combined with statements from other, independently managed Data Spaces. Tap a node to look it up.
So why no Mosaic moment? The cost of interaction.
RDF became a W3C Recommendation in 1999. Ontologies, SPARQL, RDF stores, and Linked Data publishing kept maturing. Yet mainstream adoption stayed limited next to the document Web. The architecture was powerful; using it was hard.
Four walls between people and the Web of Data
The standards existed. The implementations worked. The missing piece was an accessible, general-purpose client.
Chapter 3 · The central argument
AI agents are the Mosaic moment for hyperlinked data.
What Mosaic and Netscape did for hyperlinked documents, LLM-powered AI agents are doing right now for hyperlinked data.
The rhyme, row by row
Hover or focus a rowThe highlighted rows carry the argument: the interface is the difference, and it has arrived.
An agent can help generate RDF from a document, construct a SPARQL query, interpret ontology definitions, follow linked identifiers, and combine information from multiple Data Spaces. It can invoke tools that perform authorized operations on the underlying resources. None of this requires the LLM to become a database, a knowledge graph, or a system of record. The architecture is stronger when those responsibilities stay separate.
The symbiosis: each part keeps its job
Tap a roleThe client doesn’t replace the architecture. It unlocks its utility.
LLMs don’t make RDF or ontologies obsolete. They make them more valuable by cutting the cost of using them. A developer no longer hand-authors every RDF statement. An analyst doesn’t write every SPARQL query. A knowledge worker asks in natural language while an agent uses structured, governed context to formulate and validate answers.
An LLM on its own generates incorrect RDF, invents relationships, misuses ontologies, and writes invalid queries. That is exactly why this works: the semantic foundations keep the agent honest. Interoperable, trustworthy outcomes come from:
The barrier is gone. Ordinary users now reach semantic technologies through agent-mediated workflows, and widespread utility is delivering the vindication that architectural arguments never could.
Chapter 4 · Two journeys, one recurring pattern
History doesn’t repeat. Architecture rhymes.
Both depend on loose coupling. Both depend on open standards.
The first journey made documents universally navigable. The second is making data, information, and knowledge universally navigable, under identity, authorization, and governance.
The breakthrough that captures attention isn’t always the one that created the value. The mouse became the icon of Engelbart’s demo. The browser became the symbol of the Web. Today LLMs draw the spotlight, but their lasting value depends on what came before them: RDF, Linked Data, ontologies, knowledge graphs, databases, filesystems, APIs, and the open standards that connect them.
Coda