I’m a practicing developer and architect who has spent the last few years living at the intersection of modern web frameworks, SEO, and AI tooling. Every day, it becomes harder to pretend that the way we design interfaces can stay the same while user behavior, search, and AI are shifting under our feet. This piece is about a new kind of interface — not just another set of trendy components, but a different model of how humans interact with web applications. It’s about what happens at the crossroads of AI chat and traditional websites — and what that means for designers, developers, and businesses building products for the next 5–10 years. How we learned to use the web For the last twenty years, the web has been surprisingly predictable. There is a page. On that page, there is a header, a footer, navigation, a couple of links to neighboring pages, and sometimes a search box. Somewhere deeper live filters, categories, tags, and endless pagination. The mental model is simple: the web is a library, and every site is a small private collection with its own catalog and shelves. We learned that to reach the right “shelf”, you first have to understand how the librarian thinks. On the web, that librarian is the information architecture. You don’t just look for “something about auth”; you learn that in this product, docs live in “Documentation → API → Authentication”, while guides live somewhere else. After a few clicks and a few minutes of scrolling, you start to feel that you are “familiar” with the product. Search engines like Google and Bing amplified this model instead of replacing it. They became a global catalog on top of all those libraries. But the outcome of every search was still the same: a list of pages. We got used to googling, opening 5–10 tabs, and manually stitching together an answer from fragments scattered across different sites. It felt normal, even inevitable — that’s just how the web works, right? Google How AI chatbots broke the old mental model How AI chatbots broke the old mental model Then, large‑scale access to AI chat apps arrived. At first, they looked like toys: fun to poke at, capable of jokes, sometimes hallucinating confidently wrong things. But very quickly, something subtle but important changed — not in technology, but in how people think about asking questions. People stopped compressing their thoughts into “2–3 keywords”. Instead of typing “buy sneakers nyc”, they started writing: “I need comfortable sneakers for everyday walking, not for running, budget under \$100, okay with either NYC pickup or fast shipping.” In a traditional search engine, this kind of query feels strange. In a chat, it feels natural. And the dangerous part for the “old web” is that in this moment, the user no longer cares where the answer comes from. The cognitive model is shifting. Before, the user had to think: “How do I phrase this so the search engine understands and gives me half‑relevant links?” Now the question is: “How do I explain this the way I would to a human?” That’s the difference between “adapting to the machine” and “speaking like a person”. Chat removes a layer of technical discipline: users don’t need to remember exact page names, the right product term, or the structure of your docs. They just need to describe their situation — and if the answer is good enough, they may never visit your site at all. If AI is so smart, why do we still need websites? If you push this line of thought to the extreme, you get a radical question: if AI can answer most questions, why do we need websites at all? Maybe everything moves into one universal chat window, and pages, navigation bars, and landing layouts become museum artifacts of early web design. Technically, the answer can be almost “yes”. It is possible to imagine a world where nearly everything happens inside a chat interface: from finding products and checking out, to signing contracts and managing subscriptions. In many domains, we are already halfway there: internal support bots, scripted customer service, voice assistants that pretend to be humans on the phone. But on the level of human experience and business, the picture looks very different. A website is not just functionality. It is also a stage, with lights and sound and scenery. It is a space where a brand gets to talk in its own language — through color, composition, animation, visual metaphor. A chat is a meeting room. It’s great for clarifying, negotiating, asking quick questions. It is terrible at building atmosphere and identity. In chat, every brand looks almost the same: text bubbles, maybe an avatar, a slightly different tone of voice. For businesses, that is not just an aesthetic tragedy. It is a risk to trust, differentiation, and long‑term relationships. Visual language is a way to show that there is a real product, a real team, and a real story behind the interface. If everything collapses into a gray chat panel, all you have left is a disembodied “voice” — and it is much easier for that voice to pretend to be someone it is not. So no, pure chat will not “kill” websites. It might absorb a huge chunk of tasks that previously required navigating through pages. But it will not replace everything, because people still like to “see” a product, not just “talk” to it. Why the old page‑based web breaks in an AI world That said, the old “everything is a page” approach also fails to survive contact with reality in 2025. Think of a mature SaaS product: years of development, dozens of sections, hundreds of doc pages, blog posts, landing pages, and onboarding guides. Each piece of content made sense when it was created: “Let’s put this in a separate page so users don’t feel overwhelmed”. But from the user’s perspective, complexity accumulates. They don’t know which page holds the answer. They don’t know which of the ten similar articles is the most up-to-date. They don’t know how to connect pieces scattered across your blog, docs, and changelog. They are forced to do manual “integration testing” of your content, clicking through screens and mentally merging partial answers into something usable. AI, in this context, acts as a synthesizer. It can pull meaning from several pages and turn them into a fresh, coherent answer. Classic web UX cannot do this by design; it was built around “show this page”, not “assemble this answer”. But AI chat has a weakness too: it rarely shows the full path. It gives you the conclusion, yet rarely gives you the form — the structure, the context, the place where this lives in the system. If you extend the theater metaphor, a traditional website is the stage where you watch the whole play. An AI chat is the critic who retells the story in its own words. Sometimes that is exactly what you want; sometimes it is not. Either way, it is a different plane of experience. That tension creates a need for a hybrid interface: something that can both show and answer. The new interface: parallel experience streams This brings us to the key idea. The new interface is not “a website with a chat widget in the corner”, nor “a chat that occasionally opens webviews in a browser tab”. The new interface is a consciously designed system of several parallel experience streams that live together on one screen. One stream is conversational. This is the AI you can talk to, that understands tasks, not just URLs. It can propose paths, ask clarifying questions, and warn you before you step into a dead end. Another stream is visual and structural: pages, dashboards, tables, maps, forms — everything that requires focus, hierarchy, accessibility, and brand expression. A third stream is business logic and data: roles, permissions, constraints, workflows, and the actual state of the system. The important shift is that these streams no longer run “one after another” — first chat, then UI, then back to chat. They can and should run at the same time. The user talks to AI and simultaneously watches the interface evolve. The interface suggests something, and the user clarifies in chat what they really meant. Dialogue and visual layer stop competing for attention and start playing on the same team. Technically, this pulls us toward slot‑based layouts and parallel routes: the interface is split into independent regions, each with its own lifecycle, all coordinated by a shared scenario. Why slots and parallel routes made sense At some point, this stopped being an abstract design discussion and turned into a concrete architectural problem in one of my own projects. The requirements looked like this: Keep a product‑aware AI chat on the left, with access to internal docs and external knowledge via vector search. Show pages on the right — from static marketing content to complex authenticated UIs. Make sure any error on the right never kills the chat or resets the conversation. Preserve SEO: public content should still be delivered as static HTML, not as a JS‑dependent shell. Avoid a mess of iframes and fragile microfrontends that are painful to test and maintain.[ Keep a product‑aware AI chat on the left, with access to internal docs and external knowledge via vector search. Show pages on the right — from static marketing content to complex authenticated UIs. Make sure any error on the right never kills the chat or resets the conversation. Preserve SEO: public content should still be delivered as static HTML, not as a JS‑dependent shell. Avoid a mess of iframes and fragile microfrontends that are painful to test and maintain.[ On the architecture level, this turned into an equation with several unknowns: independence, resilience, SEO, and developer experience. In that equation, slot‑based layout (independent “windows” or slots on the screen) and parallel routing (routes that can update independently) turned out to be a natural answer. Instead of thinking in “pages”, it became more useful to think in “flows”: the left slot is the conversation flow (chat, auth, assistants), the right static slot is public content that works even with JS disabled, the right dynamic slot is personalized, authenticated functionality. From that, a new architecture emerged where AI chat and the classic site stopped fighting for control over the screen. They got their own “campus buildings”, connected by a shared campus of navigation, layout, and brand. Practically, this is what sits behind the AIFA starter templates: a Next.js‑based open‑source setup designed to keep AI chat, static SEO pages, and dynamic app surfaces in one coherent experience. What changes in real products High‑level ideas are nice, but interfaces live or die in real scenarios. Here’s how this parallel‑streams model reshapes some familiar patterns. Docs and learning products Traditional documentation is a forest of sections. Users know the answer is “somewhere in here”, but not where exactly. They skim the table of contents, try to guess by headings, open multiple tabs, and hope the right combination of pages eventually clicks. The more your product grows, the more invisible your best content becomes. In a new interface, the user starts differently: “How do I rotate an auth token in a multi‑tenant app without breaking existing sessions?” The AI layer knows the shape of your docs. It can assemble a cohesive answer from multiple pages and, if needed, open the relevant section on the right with the exact paragraph highlighted. The user sees both the synthesized answer and the “source of truth” — and can dive deeper without getting lost in the tree of pages. E‑commerce Most online stores lean heavily on filters. Filter by brand, size, price, color, and material — sometimes all at once in a dense sidebar. Very few users enjoy filling all of these out. They approximate, misclick, and then bounce when results feel slightly off. The interface is optimized for the database, not for the conversation in the buyer’s head. In a parallel‑stream setup, the user speaks first: “I’m looking for black sneakers without giant logos, for city walking, size 10, under \$100.” The chat understands that this maps to a specific category, applies filters under the hood, maybe clarifies brand preferences, and then fills the visual slot with large, clear product cards. Filters still exist — but now they are tools for refinement, not the main entry point. The user does not have to translate their intent into your filter UI; the AI layer does that translation. B2B and admin panels Complex B2B systems are notorious for steep learning curves. They have dozens of screens, each with dozens of fields, and onboarding often sounds like: “Watch these ten videos and read the docs; you’ll get used to it.” Every new customer pays the cognitive tax of understanding how your internal model maps to their real‑world tasks. With a new interface, the first step can be different. A user might say: “Show me customers whose churn increased over the last three months, but whose average contract value is still high.” The conversational layer turns this into a query over your data, opens the right report on the visual side, and explains in plain language how it interpreted the criteria. You don’t have to automate everything, but even the option to have a dialog over the interface is a qualitatively different level of experience. What this means for designers For designers, this new interface is both a challenge and a gift. The challenge is that static screen maps are no longer enough. Now the question is: what does the conversation look like? How do you visually connect a specific chat message to a change on the screen? How do you show that this particular view is “the answer” to a particular question? The gift is that you can finally stop pretending the interface is just a set of static frames. You can direct the experience like a play: there is a leading voice (the AI), there is a stage (screens and slots), and there is light and sound (animations, highlights, contextual markers). You can invent ways to visualize dialogue — without destroying structure and accessibility in the process. There is also a branding challenge: not letting your product dissolve into the same generic chat bubbles everyone else uses. Your product still needs a personality — including in the way your AI speaks. Tone of voice, microcopy, visual framing around the chat, how the interface reacts to uncertainty or errors — all of that becomes part of UX. In a world where the content layer is increasingly generated, character becomes a key differentiator. What this means for developers For developers, the new interface means the job is no longer just “build routes and components”. You have to think in terms of flows and slots. Which parts of the interface should be navigation‑independent? Which slots must survive when others crash? What is rendered statically, what dynamically, and what can be generated on demand by AI? It also means designing communication between slots. When is the chat allowed to open pages? When can a page trigger a question to the chat? How do you avoid circular dependencies and race conditions while keeping the experience seamless? Dropping a chat widget into every page is no longer enough. You have to architect the experience itself — how users move between dialogue and visual context without noticing the internal technical seams. On the technology side, this pushes you toward tools that handle slots and parallel routes well, and away from “one giant SPA that crashes all at once”. In practice, that often means leaning into frameworks like Next.js App Router, where you can define independent layouts, parallel segments, intercepting routes, and mixed static/dynamic rendering. Architectures like AIFA build on top of that: chat in one slot, public static content in another, personalized app surfaces in a third — each with its own error boundaries and lifecycle. What this means for the business For a business, the new interface is not “a fancy chat bubble on the site”. It is a way to keep control over how AI talks to your users. If you leave everything to external systems, the conversation with your customer happens in somebody else’s shell: the user types into a third‑party AI app, and that app decides which tiny fragment of your content to show or paraphrase. You are just a data source. If you embed AI into your own architecture, you get several advantages. You keep SEO traffic by serving rich static content in your own layout. You increase conversion because the path is guided by an assistant that understands your specific processes, not generic best practices. And you can build new user journeys faster by teaching the AI new concepts and language, instead of redrawing dozens of screens for every new use case. Of course, this is not free. A new interface requires investment in architecture, data quality, and conversational design. But in return, your product stops being “one more link in someone else’s search result” and becomes an environment where AI and users talk in the language of your product — on your terms, in your visual space. Risks and illusions It’s important not to turn this into yet another wave of uncritical AI hype. The new interface has traps of its own. The first illusion is believing that chat will solve everything. It won’t. Some users simply don’t like typing. Some scenarios require predictable, highly structured forms rather than open‑ended conversation. There are accessibility constraints and legal requirements that make pure chat UX risky or even unacceptable. The second risk is forgetting about transparency. If AI starts changing the interface without explaining why, users feel like they are losing control. A good new interface should reveal the links between intent and outcome: “You’re seeing this screen because you asked for this.” Users should be able to retrace steps, see what was filtered, and correct the AI when it misinterprets something. The third illusion is economic: treating AI integration as “magic cost savings”. Rebuilding architecture around AI is an investment, not a shortcut. Done poorly, it can leave you with complex, fragile code, confusing UX, and dependency on a single external provider. Done thoughtfully, it can reduce friction for users and enable new business models — but the “AI tax” is real, both technically and organizationally. Has the time come for slots and parallel routes? There is no clean “yes” or “no” answer to whether the time for this new interface has “officially” arrived. But it already feels impossible to design serious products as if AI doesn’t exist. You can’t responsibly plan a 5–10 year product roadmap and act like users haven’t learned to expect dialogue, not just navigation. Ignoring that shift won’t make it go away; it will just make your product feel oddly old, even if the tech stack is brand new. Personally, this moment feels a lot like the transition from static sites to SPAs. Back then, it looked like “just another technical trick”. It turned out to be a paradigm shift. Slot‑based architectures, parallel routes, an AI layer that lives next to content instead of sitting as a thin widget on top — all of this still feels niche today. But once you build a few real projects this way, it becomes hard to go back. The simplest practical step right now is to stop thinking in terms of “pages versus chats” and start thinking in terms of “streams that need to live together on the same screen”.
