The page you are reading is a hypermedia system… — not in spite of its simplicity, but because of it. There is no JavaScript framework managing a virtual DOM…, no client-side router, no bundler, no build step. When you click an underlined term, your browser sends an HTTP request to a server, and the server sends back a fragment of HTML that gets inserted directly into this document. That is the entire mechanism. It is, in the language of Roy Fielding…, HATEOAS….
The Web's Forgotten Architecture
Carson Gross's book Hypermedia Systems… opens with a provocation: HTML, the world's most widely used hypermedia, is today viewed resentfully by many of the developers who build with it — an awkward legacy markup language that happens to be there, in the browser, and so must be grudgingly used. The book's argument is that this is a profound misreading. HTML is not a limitation to be worked around. It is a complete application architecture… — one that the industry largely abandoned in the mid-2000s without fully exploring what it could do.
The abandonment happened gradually, then all at once. Single Page Applications… arrived with the promise of desktop-like interactivity, and delivered it — at the cost of downloading a JavaScript runtime, constructing a parallel model of the page in memory, and shipping application logic to the client that the server used to handle. The web became, in effect, two systems pretending to be one: a thin HTML shell delivering a JavaScript application that then rebuilt the interface from scratch.
What the Hyperlink Actually Is
Before we had hyperlinks, we had footnotes and commonplace books… — physical implementations of the same idea: traversable paths between related concepts. The Talmud… understood non-linear reading centuries before Ted Nelson coined the word "hypertext" in 1963. What the web promised was to remove the friction from that practice — to make the associative trails that Vannevar Bush imagined… in 1945 instantaneous and global.
Instead, the hyperlink got domesticated. It migrated from prose into navigation bars, sidebars, and footer link farms. The link became interface furniture… rather than a literary device. And then SEO colonized it…. The hyperlink — this technology that was supposed to create permanent, traversable networks of meaning — became fragile, manipulable, and ultimately unreliable.
How This Page Is Built
This article is built with HTMX…, a library whose central argument is that HTML's native hypermedia controls… — the anchor tag and the form tag — are arbitrarily limited. Why should only anchor tags make requests? Why only on click? Why only GET and POST? Why should the response always replace the entire screen? HTMX removes those constraints, giving any HTML element the ability to issue any HTTP request and swap the response into any part of the document.
Every expandable term in this article is an ordinary anchor tag carrying a handful of hx- attributes…. When you click, it posts to a PHP endpoint and inserts the returned HTML fragment directly into this document — no page reload, no state reconciliation, no virtual anything. The document's state lives in the document. That is what Hypermedia As The Engine of Application State… actually means in practice.
What This Means for Coding Agents
There is a second argument for this architecture that has nothing to do with philosophy and everything to do with the practical limits of
coding agents….
These agents struggle not just because codebases are large, but because so much of what fills their context windows is
semantically empty syntax…
— framework boilerplate, build configurations, JSX transformations, and utility-class markup that tells an agent nothing about what anything means. An agent reading <div class="flex items-center justify-between px-4 py-2 border-b"> knows as much about the purpose of that element as you do after staring at a foreign language menu: it can follow the grammar without understanding a word.
Semantic HTML is the opposite.
Semantic elements…
— <article>, <nav>, <footer>, <time>, <aside> — tell an agent what something is, not just how it looks. A hypermedia-driven architecture compounds the benefit: less framework surface area means smaller context, more signal per token, less
synthetic abstraction…
to reason through. And at training scale, thousands of well-structured hypermedia pages give models better signal than thousands of pages of
div soup….
The semantic web was supposed to make the web legible to machines. We never quite got there — but we did not have to abandon the attempt entirely.
Which brings the argument full circle. The same PHP endpoint that serves HTML fragments to your browser via HTMX could serve structured responses to an AI agent via WebMCP…. Same resources, same relationships, different representation. A browser gets HTML; an agent gets a tool result. One backend, two clients. The server knows what things mean — it has always known — and both kinds of reader benefit from that knowledge being encoded in the structure of the document rather than buried in a JavaScript bundle.
Where the Hyperarticle Points
Each link in this document also carries coordinate metadata… — its paragraph, sentence, and link index in the original document. Right now, the server uses those coordinates to look up a pre-written response. The next version hands them to a language model, which generates a coherent expansion in context — aware not just of the term clicked, but of where it sits in the argument.
That extension — from static lookup to live generation — maps directly onto the data-lm-prompt proposal… outlined in the Banapana essay this article grew from: extend the anchor tag with a query attribute, let the browser resolve it dynamically, keep the static href as a fallback. The hyperarticle is that proposal made concrete — not waiting for browsers, not requiring a spec, running today on a PHP server and fourteen kilobytes of JavaScript. The link is not a button. It never was. And the document is not finished when the author stops writing.