The web development world spent the better part of a decade convincing itself that the browser should do the heavy lifting. We started with documents and forms, then steadily migrated every conceivable operation to the client. Routing, state management, data fetching, rendering logic, even database query orchestration through GraphQL — it all moved into JavaScript bundles that grew heavier with each release. Frameworks multiplied, build pipelines deepened, and what began as a way to make apps feel snappy turned into an architecture where a page could not render a single meaningful pixel until megabytes of code had been downloaded, parsed, and executed.
That shift solved real problems. Server-rendered pages with a sprinkle of jQuery struggled to deliver the smooth, app-like transitions users expected. Single Page Applications gave us instant navigation, persistent state, and rich interactions. But the cost accumulated. Teams now manage complex client-side state stores, wrangle massive JavaScript bundles, maintain finicky data synchronization layers, and debug build pipelines that sometimes feel like their own full-time job. We traded one set of problems for another, and many developers are now asking whether every application needs to pay that tax.
Two developments are making that question easier to answer.
HTMX and the Return of Hypermedia
The first is HTMX. On the surface, it looks like a small library, but its architectural implication is large. HTMX treats HTML as the native format for application logic instead of treating it as a static shell that must be inflated by JavaScript.
Here is what changes in practice. Traditionally, when a user clicks a button to load more comments, the frontend fires a fetch request, receives a JSON payload, normalizes it into a client-side store, runs it through a component template, diffsthe virtual DOM, and finally patches the page. HTMX short-circuits that chain. The button itself contains attributes that tell the browser where to send the request and which page element to replace. The server returns an HTML fragment — just the new comments, wrapped in a div. The browser swaps it in. There is no JSON, no frontend state tree, no reconciliation algorithm, and no imperative JavaScript to keep the UI in sync with the server.
This is not a rejection of modern development. It is a rejection of unnecessary abstraction. HTMX proves that hypermedia, the architectural style that powered the early web, can still support sophisticated interfaces when paired with modern ergonomics. Any element can issue requests, not just forms and links. Any event can trigger an update. The server remains the source of truth for both data and presentation.
Chrome’s Declarative Partial Updates
The second shift is newer and lives inside the browser itself. Chrome is introducing Declarative Partial Updates, or DPU. This feature allows a browser to stream HTML and insert it directly into targeted parts of a page as the bytes arrive.
Before DPU, if you wanted to stream live data into a web page, you generally reached for WebSockets, Server-Sent Events, or long-polling paired with manual DOM manipulation. The frontend had to manage the connection, parse the payload, and decide exactly how and where to inject markup. DPU changes the equation by making the process declarative. The developer specifies a target container, and the browser handles the rest: receiving the stream, parsing the fragment, and placing it exactly where it belongs, even before the full response closes.
Think of a monitoring dashboard showing server logs or a support queue updating in real time. With DPU, the backend sends plain HTML chunks as they are generated. The browser streams them into a table body or a feed container without a single line of client-side streaming logic. The assembly happens natively.
The Server-First Model
Put HTMX and DPU together, and you get a coherent architecture where the server owns the state and generates the UI, while the browser handles display and user input. Backend frameworks like Rails, Laravel, Django, Go templates, or ASP.NET become the primary interface layer again. The frontend is not a separate application that consumes an API. It is the hypermedia interface the server produces.
This model fits a surprisingly wide slice of software. Consider the typical SaaS application. It is dashboards with sortable tables. It is admin panels with forms and filters. It is internal tools that move records from one state to another. It is CRUD workflows that show a list, expose a detail view, and let the user edit fields. It is even AI interfaces where a language model streams tokens back to the user, and each token or paragraph can be wrapped in HTML and appended to the conversation thread. For all of these, a thick JavaScript client is often overkill.
The benefits are immediate and practical. Initial page loads are faster because the first meaningful paint arrives as HTML, not after a hydration cycle completes. JavaScript payloads shrink because there is no virtual DOM, no client-side router, and no state management library to ship. Search engines see complete content without executing bundles, so SEO works by default. Complexity drops because one codebase handles routing, business logic, and rendering. Debugging gets easier. When something looks wrong, you inspect the Network tab and see exactly what HTML the server sent. There is no opaque client-side state object to reverse-engineer.
What About React and Heavy Clients?
None of this means React is dead, or that SPAs are a mistake. Complex, editor-grade applications still need a thick client. Figma runs a C++ engine compiled to WebAssembly inside the browser because server round-trips would make drawing impossible. Canva manipulates the canvas at sixty frames per second with client-side geometry. Google Docs uses operational transforms to resolve editing conflicts in milliseconds. These tools are essentially desktop applications delivered through a browser tab. They are not going back to server-rendered forms.
But most software is not Figma. Most software is not a real-time graphics editor. Most software is a reporting screen, a configuration panel, a booking flow, or a content management form. For that long tail of applications, shipping hundreds of kilobytes of JavaScript framework just to toggle a modal or fetch a list of records never made much sense. The economics of the stack are shifting. We are rediscovering that the server can be close to the user thanks to the edge, and that the browser itself has grown capable enough to update fragments without a framework intermediating every byte.
The Pendulum Finds a Balance
The arc of web architecture is swinging back toward simplicity, but it is not a naive return to the nineties. The browser is becoming smarter. Features like DPU do not replace developer ingenuity; they absorb the patterns we used to implement by hand — streaming, partial updates, targeted DOM insertion — into the platform itself. HTMX gives us the vocabulary to express those behaviors without reconstructing a miniature operating system in the frontend.
You no longer have to choose between a simple architecture and a responsive user experience. You can have both. The server can drive the interface, the browser can assemble it, and the JavaScript you write can focus on genuine interactivity rather than plumbing.
For the next generation of dashboards, admin tools, and AI-powered interfaces, the smartest client might be the one that does less.