Inside the den · The thesis

Why Chat-Based Websites Are the Future of the Web

The web was built for documents. We’ve spent thirty years pretending it was built for people. Conversational interfaces finally close that gap.

Start Speaking Free About FluentFox

The premise

From brochure to instrument.

The page-and-navbar web is an artifact of 1990s technology, not a law of nature. It was right when the web was a library of documents. It is wrong now that the web is where we do things.

State your intent. Let the system assemble the response.

A chat-based interface (“ChatUI”) inverts the model. Instead of the user navigating a tree of pages to find the one screen that does what they want, the user simply states their intent, and the system assembles the response. The burden of navigation shifts from the human to the machine — which is exactly where it belongs.

But the change runs deeper than the interface. It changes what a website is. The page-and-navbar site is, at heart, a brochure — inherited from print, arranged to present and persuade, something you visit and browse. A conversational site is an instrument — a tool you pick up and use, that asks “how can I help?” and gets out of your way.

So this article argues that conversational, intent-driven interfaces will become the primary surface of the web, that the traditional page/link/navbar model will recede into a supporting role, and that the sites which win will be the ones that stop behaving like brochures and start behaving like instruments.

The core idea

A store with no staff.

Walk into a good store and there are two things you can do.

Aisles and staff. Nobody designs a store with only one.

You can browse on your own — wander the aisles, read the shelves, see what catches your eye. There is nothing wrong with this. Sometimes the wandering is the pleasure. Browsing is a legitimate, often delightful, way to shop.

Or you can ask a salesperson: “Where are the teas?” — and be walked straight to the right shelf. You said what you wanted, and someone who knows the store took you there.

Every good physical store understood this centuries ago: it has both. The shelves are beautifully arranged for the browsers, and there’s always someone to ask for the people who just want their tea.

The web spent thirty years building only half of it. We built the aisles — gorgeous, elaborate aisles. Pages, categories, mega-menus, a navigation bar that is really just aisle signage hung from the ceiling. And we staffed them with no one. For three decades the web has been a store with no staff: every shelf in its place, and not a soul to ask.

A conversational interface is how the store finally hires the clerk. It does not bulldoze the aisles — it adds the missing half: someone to ask.
— The whole argument, in one line

01 · The tax

Navigation is a tax, not a feature.

Every navigation bar is a confession: “We couldn’t figure out what you wanted, so here are forty places you might find it.”

Five steps of overhead, collapsed into one.

Consider what a user actually goes through on a traditional site:

Form an intent (“I want to change my billing address”).
Translate that intent into the site’s information architecture (“Is that under Account? Settings? Billing?”).
Scan a menu, guess, click.
Land on the wrong page. Backtrack.
Repeat until found.

Steps 2–5 are pure overhead. They exist only because the interface cannot understand intent — so the user must learn the map the designer built instead of just saying what they mean. A chat interface collapses all of this into step 1.

The whole interaction, in two lines

User: Change my billing address to 14 Oak Street.

System: Done — updated your billing address to 14 Oak Street. Want me to update shipping too?

The best navigation is no navigation.
— There is no menu structure to learn

02 · Intent

Intent is the real unit of interaction.

Pages model the system’s structure. Conversations model the user’s goal. These are rarely the same shape.

One sentence, and the orchestration happens behind the curtain.

A single user intent — “plan a weekend trip to Lisbon under $800” — might touch flights, hotels, a calendar, a weather service, and a budget tracker. In a traditional web app that’s five tabs, five logins, and a spreadsheet to reconcile them. In a conversational interface it’s one sentence.

Even the humblest task shows the shift. Take buying a pen. The old modality — type “pen” into the search box, get back 4,000 results, and now you do the work: filter by price, wade through ballpoints and fountain pens and gel pens, sort, compare, scroll, second-guess, click. The search box took your intent and handed it straight back to you as a wall of options to sift.

The new modality — state intent, get a dialogue

User: I want to buy a pen.

System: What kind — ballpoint, gel, or fountain? And what price range?

User: A nice gel pen, under $15.

System: Here are the three best-reviewed gel pens under $15. Want them in black, or a color?

Let the system decompose the goal.

The system asks the one or two questions that actually matter and converges on the answer in a couple of turns — the way a good shop clerk would. Search makes you guess the right keywords and then drown in results; conversation lets you say what you mean and be asked the rest.

This is the deeper point: the page-based web forces users to decompose their goals into the system’s vocabulary. The conversational web lets the system decompose the goal instead. Decomposition is hard cognitive work, and we’ve been outsourcing it to users for decades because we had no alternative. Now we do.

03 · Generative UI

The interface can finally be generated, not pre-built.

For thirty years, every screen a user might see had to be designed and built in advance. Modern language models break this constraint.

The interface is a function of intent, not a fixed artifact.

The interface can be generated on demand, shaped to the specific request:

Ask for a comparison, and you get a table — rendered just for that question.
Ask for a trend, and you get a chart.
Ask to complete a task, and you get exactly the three inputs that task needs, and nothing else.

The chat thread becomes a container into which the system streams whatever widget, form, visualization, or block of prose best answers the moment. The navbar offered N destinations. Generative chat offers infinite ones, each built precisely when needed and discarded when done.

And because the interface is generated rather than pre-drawn, it can fit the exact device it lands on. The same intent yields a wide two-column layout on a desktop, a single full-bleed card on a phone, a spoken sentence on an earbud — not three responsive breakpoints grafted onto one rigid template, but three genuinely different renderings of the same underlying meaning.

04 · Portability

One modality, every device, every ability.

A conversation is the most portable interface humanity has.

No layout to break.

It works:

By voice, hands-free, while driving or cooking.
By text, silently, in a meeting.
On a watch, where a navbar is absurd.
For screen-reader users, for whom a linear conversational flow is native rather than a retrofit.
Across languages, because translation is a first-class capability of the underlying model, not a localization project.

The page/navbar paradigm fractures across screen sizes — the responsive-design industry exists entirely to paper over this. A conversation has no layout to break. It scales from a phone to a kiosk to an earbud without redesign, because its fundamental unit is meaning, not pixels.

05 · Memory

Memory and context as default, not afterthought.

Traditional sites are amnesiacs. Each page load is a blank slate; “personalization” is a bolt-on that tracks you to sell you things.

The interaction accumulates value instead of resetting on every click.

A conversational interface is stateful by nature. It remembers what you said three turns ago, last week, last month. It can carry your preferences, your history, and your half-finished tasks forward without you re-entering anything.

“Same as last time, but make it vegetarian.”
— Impossible in a page-based world without a dozen forms. The natural way to speak in a conversational one.

06 · Case study

FluentFox on a phone.

Language learning is the proof. It is conversational at its core — you learn a language by talking — and it is the kind of experience that suffers most from the page/navbar model.

The entire experience is the conversation.

Duolingo-style apps are a grid of menus, lesson tiles, progress screens, and settings tabs wrapped around what should be a single continuous dialogue. The navigation is the friction. FluentFox is built the other way around. There is no lesson menu to browse, no tab bar to thumb through — the engine plans a lesson, then feeds the learner one exchange at a time, and the thread simply scrolls forward.

The render signal is already in the data. Each exchange the engine produces carries its own ExchangeRender hint in the catalog — the engine doesn’t just say what to teach, it says how this turn should appear. That is generative UI by another name: the server emits intent-plus-presentation, and the client renders the precise widget for the moment:

IntroduceVocabulary → a clean card with the word, its meaning, and a play button. Pure input, nothing to tap but Continue.
RepeatAfterMe / ReadAloud → a microphone front-and-center, the target phrase above it, a waveform while you speak.
NameTheImage → an image fills the screen and the mic waits — no text, no chrome.
FillInTheBlank → the sentence with a gap, and just the controls that gap needs.

The learner never navigates to any of these. Each one is summoned by the lesson plan, rendered, scored, and discarded.

The phone is the best case, not the hard case.

A traditional learning app fights the small screen — menus collapse into hamburgers, dashboards shrink into unreadable cards, the navbar eats vertical space that’s already scarce. FluentFox has none of that to compromise. On a phone the conversational model is better than on desktop:

One exchange owns the whole screen. The current turn is the entire viewport, which is precisely what a small screen wants.
Speech-first means no keyboard. FluentFox’s exchanges are spoken, so the cramped on-screen keyboard never appears. The mic is the primary control, and a mic is easier on a phone than on a laptop.
Big tap targets, thumb-reachable. The few controls that exist — Continue, a mic, an image — can be large and sit in the thumb zone.
Hands-and-eyes-free is in reach. Because the interaction is fundamentally voice, a learner can practice while walking or commuting.

FluentFox isn’t adapted to mobile; the conversational model is mobile-native, and the page/navbar app is the thing that has to struggle to get there.

07 · The honest objections

The counterarguments — and why they soften.

A fair article names the strongest objections. Here are the real ones, and where they land.

“Discovery suffers — how do I learn what’s possible?”

Real concern. A blank chat box is intimidating in a way a menu is not. The answer is suggested actions, examples, and progressive disclosure — the system proposes what it can do, conversationally, the way a good salesperson does. Discovery moves from a static menu to a guided dialogue. This is a design problem, not a dead end.

“Some tasks are inherently spatial.”

Editing a photo, arranging a seating chart, reading a dense dashboard — these want direct manipulation, not prose. Correct. The future isn’t chat replaces everything; it’s chat is the front door, and it summons the right tool. You ask, and the spatial tool appears inside the conversation. Generative UI means the canvas and the conversation coexist.

“Conversation is slow for repetitive tasks.”

Typing a sentence is slower than clicking a bookmarked button you press daily. True — for high-frequency, muscle-memory tasks. The conversational interface should learn these and offer one-tap shortcuts. The two models blend; they don’t fight.

The honest conclusion is not that links and pages vanish. It’s that they stop being the primary organizing principle and become components the conversation calls upon.

08 · The reframe

From brochure to instrument.

Step back, and the deepest change isn’t to the interface — it’s to what a website is.

The brochure

Something you visit

You go there to look, browse, be persuaded
Organized around the company's message
Success looks like impressions, clicks, time-on-page
Inherited from print and advertising

The instrument

Something you use

You go there to act, accomplish, get an outcome
Organized around your intent
Success looks like the task done, then you leave
Inherited from the conversation and the tool

Time-on-site — the brochure’s proudest metric — becomes a sign of failure.

A brochure wants your attention and measures itself by how long it holds you. An instrument wants to get out of your way and measures itself by how quickly it gets you the result. For an instrument, a long visit means the tool was slow.

FluentFox is squarely an instrument. No one opens it to admire a landing page; they open it to speak a sentence and learn a word. The conversational model isn’t a coat of paint on a brochure — it’s what lets the product stop being a brochure at all.

09 · The front door

FluentFox’s front door: the Personless Salesperson.

The store metaphor stops being a metaphor here.

A prospect at the threshold, with a shopper’s questions.

Picture the actual visitor. Someone said “check out FluentFox.ai,” and here he is. He is not here to do a lesson — he hasn’t decided anything yet. He’s a prospect, sizing the place up: Is this for me? Does it do my language? How is it different from the app I already quit? What does it cost? These are a shopper’s questions, not a learner’s.

So give him the same two doors a real store gives him. Land him on a page split down the middle — an explicit, honest choice, neither option hidden. On one side, the clerk: “Ask me anything about FluentFox.” A single input, speak or type. On the other side, the aisles: “I’d rather just look around” — one click and the whole traditional site is his.

If he picks the clerk, he gets a salesperson — not a lesson

Prospect: How is this different from Duolingo?

FluentFox: Duolingo has you tapping word tiles; here you speak from the first minute and I listen and correct you — like a tutor, not a quiz. Want to hear what that’s like? Say this with me: “Un caffè, per favore.”

Prospect: Do you do Japanese? And what does it cost?

FluentFox: Japanese, yes — including the speaking. There’s a free tier to start with no card, and unlimited is $12 a month. Want me to start you in Japanese right now?

The clerk answers the question first — then offers a taste, then the door.

The pitch, the demo, and the sign-up are all just places this one conversation can go. He never had to find the “Pricing” page or the “Languages” page; he asked, and was taken there. That’s the clerk walking him to the tea.

If he picks the aisles, the clerk goes into hiding. The conversational panel collapses to a small, persistent “Ask a question” affordance in the corner, and the full traditional site is his to wander — exactly as it is today. The clerk doesn’t vanish; it steps back and stays within reach, the way good staff hover near enough to ask but far enough not to hover.

Why both doors — and not just the clerk?

Because a chat box can also scare a visitor off. Someone who has been burned by a clumsy chatbot — the support bot that loops, the assistant that never quite understands — flinches at a chat box the same way they flinch at a pushy, unhelpful salesperson. That memory is real, and it is exactly why we keep the user’s options open: the aisles are always one click away, so no one is ever trapped in a conversation that isn’t working for them.

But the bad chatbot is a stage, not a verdict. One day, when the conversational experience is genuinely perfected, the chat UI won’t be one of two doors — in an AI era, it will be the only one anyone reaches for.

FluentFox already replaced the tutor with software that teaches like one — the Tutorless Tutor. The front door replaces the salesperson with software that sells like one — the Personless Salesperson.
— Same move, same medium, opposite ends of the journey

And keep the brochure pages — for the robots, and for the browsers.

Two audiences justify keeping every existing page:

Search crawlers read pages, not conversations. The long-tail content that gets FluentFox found stays exactly as it is — it’s the discovery layer that brings the prospect to the door in the first place.
Human browsers who pick “let me look around” need real aisles to walk. The pages aren’t legacy debt; they’re the half of the store that serves the people who’d rather not ask.

So nothing gets torn down. The brochure keeps doing what it’s good at — being found, and rewarding the wanderers — and the clerk is added beside it for everyone who’d rather just ask. A store with staff, at last.

10 · The transition

What the transition looks like.

This won’t happen as a big bang. It will happen the way mobile happened — feature by feature, site by site.

Bolted on

Chat as an assistant bolted onto a normal site. Where we are now.

Primary surface

Chat as the primary surface, with pages summoned as needed. Where the leaders are heading.

Chat-native

Products with no traditional navigation at all, where the entire experience is the dialogue.

“Which page is that on?” will sound as dated as “which floppy disk is that on?”

Each step moves more of the navigation burden off the user and onto the system. Each step makes the underlying capabilities more composable. The endpoint is a web where you describe what you want and it happens.

The old advice doesn’t get refuted — it gets orphaned.

This is how a paradigm usually dies. The experts keep being right. A web designer’s counsel — restraint with animation, clear hierarchy, fast load times — stays exactly as sound as it ever was. What changes is the ground underneath: the page they’re perfecting quietly stops being how most people get where they’re going. The craft doesn’t lose its skill; it loses its audience. It survives as a specialty — the equestrian class of the web, practiced beautifully by a few — while everyone else just asks, and is taken there.

The advice for riding a horse is as true as ever. There just aren’t many horses on the road.
— How a paradigm dies: not refuted, just orphaned

The chat box is where that conversation begins.

The web's first thirty years taught humans to think like machines so machines could serve them. The next thirty will teach machines to think like humans. FluentFox is built for that web — start with a sentence.

Start Speaking Free About FluentFox