> [Given] a method for chat applications to understand components exposed from the website in question [...] AI chat applications win, and so does the brand that gets to keep ownership of how it tells the story of its product to its users.
Never happening. The brand may "keep ownership of how it tells its story," but it loses its users. You have turned your tool into a series of widget in someone else's application, with no control whatsoever over how users interact with you. Want to show your user a notification? (Sure you do—I can't get away from the things.) Too bad. ChatGPT owns your users, and they only see what OpenAI wants them to see, which likely will not include ads for your premium features.
Don't mistake this for user freedom, either. Users still won't own their own tools. We're just moving from a model where each vendor separately leases you their tool to a model where every tool is leased via OpenAI, which curates them based on its own monopolistic whims.
Tech companies will not surrender control of their users so easily. They may integrate chatbot components into their apps, but they will not permit an inversion of control where their product becomes a component in a chatbot.
> You likely won’t expect users 5 years from today to navigate 5 pages deep
Of course I do. There's this fallacy that, because chatbots are useful for some things, chatbot interfaces must be the best at everything, and that's just not true. I don't go to ChatGPT to ask it for relevant tech news. I go here and browse the HN frontpage. Chatbots offer zero discoverability; search bars didn't replace page navigation, and chat bots won't either.
The idea of representing UI as state goes back forever, I’m not that old but at least in the advent of the web, plenty of JSON -> UI specs or libraries have came into existence. If the specification is solid and a large portion of people agree upon it, I don’t doubt it will take over what we think of UI. (current contenders are json_render, a2ui etc)
The first benefit being that if I can describe my entire UI and the actions each component wants as json, theoretically, I can pass that file to any client to render be it mobile, react, a java swing app etc The responsibility of rendering becomes of anyone who wants to do it.
UI JSON -> UI Framework <- Design Tokens
Above is a simple way of describing how it would generally work. Where the UI framework can be whatever it wants to be as long as it knows how to connect up the UI JSON in a meaningful way.
Now for existing apps and their respective UI’s it’s never made all that much sense to describe how your components behavior in state, useful for some, and many have done it, but a hard pitch for others.
In the agentic era, the pitch is a lot more appealing.
- LLM’s are great-enough at writing JSON
- A lot of people are sharing the sentiment that they can just vibe code small apps for themselves. Hinting at they love the actual ability for full personalization.
Though having the user generate HTML and the rest all the time by LLM’s is more error prone, slow and costly.
The user can just ask an LLM to compose a composition of components in JSON laid out how they want and connected to the API’s they care about. (that can be rendered anywhere)
Personally, if I had a catalogue of 100 distinct services/API's, and I could ask an LLM to generate a UI in JSON that I can copy and paste anywhere to render it, I would be in heaven.
I think that the inverse thesis is true, you make websites more accessible (a11y, wgac, aria labels etc) for humans, then the interaction heuristics are clearer for agents functioning off browser-use or similar. If a screen reader can understand your site, then an agent can. Reinventing the wheel to facilitate the current state of agents makes the web worse for everyone, it's not a preemptive move, it's actually a decline in almost objective and measurable quality, and potentially one which removes access to the internet by people who just want to.. use the internet.
This. The accessibility tree is a superpower for agents when it's good. Screenshots are "robust" but low performance. Like so many other things, making stuff better for humans indirectly makes it better for agents.
AI component libraries in your site make your web app even more easily consumed and subsumed by AI chat clients.
This not only kills pages, but it kills the concept of a browser where the user agent is a human, rather than making your pages be designed where the user agent is an AI agent.
That doesn't make me happy to experience because I'm guessing that after a generation or so, web designers will not only do mobile first designs with stupid amounts of white space and not taking advantage of the desktops greater screen real estate and precise mouse movements, but AI first websites will get so popular that browsing sites manually will look like trying to use a text only browser in the JavaScript world.
Easy for me to do a depressing take, but hopefully the bitter lesson of AI will help this particular projected future not come to pass because the AI will get smart enough that it will embed a browser right there in line and just render the window for the user, or it will otherwise gets good enough at screen scraping and UI automation that it can just use an existing browser, just like a human, the sites won't be dumbed down even further for AI consumption.
at best we have walled garden content; and when those are scraped (either by the host or by more sophisticated bots) those walled gardens will hopefully rot under an inability to drive advertisement revenue.
I agree, I think we're at the edge of a paradigmatic shift away from humans navigating TCP-IP itself. What that looks like, I don't know, but given trends (like dynamic pricing, human-futures marketing, surveillance, and consolidation of computing under mega-companies) I can imagine: local beacons screaming AI advertisement components across a geospatial sneakernet. Auditorium-based ticketed podcasting and AR/VR/meatspace events. Thoughtful hackers reminiscing of better times simulating them in web-assembly driven first-person POV "sites" and a rolling set of encryption keys for read-access (just send them BTC)
without an ecosystem for humans to contribute meaningfully to a feedback loop that allows for free group assembly around like interests, monetary growth for hosts and other participants, and some degree of presence / searchability / permanence, the current text-only web page paradigm is doomed.
> In a world where we can type anything into a text box and get the information back instantly we are circumventing the need to visit websites altogether.
This is purely anecdotal, but the only people in my extended circle making this transition (to any extent) are the technically savvy; everyone else is slowly realizing how awful AI tools and "AI-first experiences" can be and are actively trying to avoid them.
I've noticed this bimodal distribution of perception too, and my hypothesis is that's it's hugely driven by the difference of "who is in the driver's seat".
Your tech-savvy AI early adopters are discerning between tools, the deployments and environments, and are willing and able to change things to extract the highest output from current capabilities. For instance, re-architecting a codebase to make it easier for agents to contribute to it.
The rest are having AI hypeware shoved upon them, often as a cost cutting measure, and lack the agency to influence outcomes. When agents misbehave, they only have the option to "Press 0 to speak with a Human" and hope that works.
I suspect this is a big factor in the divide we're seeing, and might result in your median adult being ambushed by recent gains in capabilities.
I think the more realistic direction is exposing API / MCP-style interfaces for agents to interact with a product’s functionality, rather than shipping UI components that an AI client would render.
The "AI renders your components inside chat" idea feels very similar to Facebook’s old canvas apps. That model disappeared for good reasons: abuse, security, and loss of platform control.
It seems far more likely that AI platforms will provide their own interaction primitives (forms, pickers, confirmations, etc.) and simply call third-party tools behind the scenes. That lets the platform retain control over UX and safety, and avoids the risks of embedding arbitrary third-party UI.
Theres probably a middle ground where we allow an app to augment / enhance itself using deterministic behaviour but a user based soft request. If I as a user can ask for a feature, and have it work just for me, thats pretty cool.
Instead of progressive enhancement it can be progressive evolution.
When I use ChatGPT to do research, I expect it to justify itself by quoting from web pages and linking to those web pages. (I gave it explicit instructions to quote things, but unfortunately it will only do short quotes.)
This might be an extended web search, but it's still a web search. The documents need to exist. Maybe a lot of the surrounding boilerplate disappears, though?
> In a world where we can type anything into a text box and get the information back instantly we are circumventing the need to visit websites altogether
But without visits many websites will disappear.
So where will AI get its information from in the future?
I do like that idea of allowing flexible component rendering, especially if you're building a your own app with chat UI. The one problem would be like always standardisation, will chatUIs need to be like browser and follow standards? Or do they need to render JS, CSS, HTML as a component? What freedom do chatUIs allow the components?
Even so, ask, what's the end goal of the user? Does it even make sense to worry about UI if we're think autonomous agents that sole goal is to accomplish something defined by the user?
> Imagine your own product. You likely won’t expect users 5 years from today to navigate 5 pages deep, apply filters and sort data just to try to derive the answer themselves… will you?
Yes, I do, because outside of tech circles, AI adoption is not as pervasive/ universal as the loudest voices keep suggesting. I’m not dismissing its value, but gentle reminder to spend more time around nontechnical people, listen to how they feel about the tech, and pay attention to how they engage with it when offered.
I like the direction this is taking, and I think the more radical and more coherent endgame is data ownership returning to us and AI's replacing website brands all together as the AI generates any UI we could ever want.
Today most websites offer interfaces for accessing the data that they hold hostage. However they are holding our data. We would prefer to have it back if we could, and AI will enable that. Our paradigm will shift this way eventually I hope.
Though it's true companies dont have an incentive to hand over data custody anytime soon, AI companies have already found a way around this in little ways and its a matter of time until it goes to where all the data we could ever want is within the AI cloud and our next task is simply taking back ownership of that (where local inference takes us).
Sadly though maybe the AI companies and our leaders-acting-as-overlords have already guessed this end game and have started to push these "subscription to laptops" and "own nothing" models with shortage of hardware and rising costs of things like Mac mini's indicating that could be a reality we experience.
Would you rather visit a website full of ads that doesn't answer your question and is written for SEO or get the answer instantly without any ads?
Eventually, after killing several websites by depriving them of revenue, ChatGPT will enshittify like everything else and starting adding ads.
There isn't even a question about that. Just think of Google for example.
Why Google's SERP has ads but Gemini does not? There isn't even a "they are making money with the data" argument here, because Google already has all the query data it could ever want. They just haven't added ads yet.
Eventually, Gemini will look like a SERP with 5 ad results, if it doesn't go to the Google graveyard like everything else.
> [Given] a method for chat applications to understand components exposed from the website in question [...] AI chat applications win, and so does the brand that gets to keep ownership of how it tells the story of its product to its users.
Never happening. The brand may "keep ownership of how it tells its story," but it loses its users. You have turned your tool into a series of widget in someone else's application, with no control whatsoever over how users interact with you. Want to show your user a notification? (Sure you do—I can't get away from the things.) Too bad. ChatGPT owns your users, and they only see what OpenAI wants them to see, which likely will not include ads for your premium features.
Don't mistake this for user freedom, either. Users still won't own their own tools. We're just moving from a model where each vendor separately leases you their tool to a model where every tool is leased via OpenAI, which curates them based on its own monopolistic whims.
Tech companies will not surrender control of their users so easily. They may integrate chatbot components into their apps, but they will not permit an inversion of control where their product becomes a component in a chatbot.
> You likely won’t expect users 5 years from today to navigate 5 pages deep
Of course I do. There's this fallacy that, because chatbots are useful for some things, chatbot interfaces must be the best at everything, and that's just not true. I don't go to ChatGPT to ask it for relevant tech news. I go here and browse the HN frontpage. Chatbots offer zero discoverability; search bars didn't replace page navigation, and chat bots won't either.
I’m interested in this space my thoughts so far;
The idea of representing UI as state goes back forever, I’m not that old but at least in the advent of the web, plenty of JSON -> UI specs or libraries have came into existence. If the specification is solid and a large portion of people agree upon it, I don’t doubt it will take over what we think of UI. (current contenders are json_render, a2ui etc)
The first benefit being that if I can describe my entire UI and the actions each component wants as json, theoretically, I can pass that file to any client to render be it mobile, react, a java swing app etc The responsibility of rendering becomes of anyone who wants to do it.
UI JSON -> UI Framework <- Design Tokens
Above is a simple way of describing how it would generally work. Where the UI framework can be whatever it wants to be as long as it knows how to connect up the UI JSON in a meaningful way.
Now for existing apps and their respective UI’s it’s never made all that much sense to describe how your components behavior in state, useful for some, and many have done it, but a hard pitch for others.
In the agentic era, the pitch is a lot more appealing.
- LLM’s are great-enough at writing JSON
- A lot of people are sharing the sentiment that they can just vibe code small apps for themselves. Hinting at they love the actual ability for full personalization.
Though having the user generate HTML and the rest all the time by LLM’s is more error prone, slow and costly.
The user can just ask an LLM to compose a composition of components in JSON laid out how they want and connected to the API’s they care about. (that can be rendered anywhere)
Personally, if I had a catalogue of 100 distinct services/API's, and I could ask an LLM to generate a UI in JSON that I can copy and paste anywhere to render it, I would be in heaven.
I think that the inverse thesis is true, you make websites more accessible (a11y, wgac, aria labels etc) for humans, then the interaction heuristics are clearer for agents functioning off browser-use or similar. If a screen reader can understand your site, then an agent can. Reinventing the wheel to facilitate the current state of agents makes the web worse for everyone, it's not a preemptive move, it's actually a decline in almost objective and measurable quality, and potentially one which removes access to the internet by people who just want to.. use the internet.
This. The accessibility tree is a superpower for agents when it's good. Screenshots are "robust" but low performance. Like so many other things, making stuff better for humans indirectly makes it better for agents.
AI component libraries in your site make your web app even more easily consumed and subsumed by AI chat clients.
This not only kills pages, but it kills the concept of a browser where the user agent is a human, rather than making your pages be designed where the user agent is an AI agent.
That doesn't make me happy to experience because I'm guessing that after a generation or so, web designers will not only do mobile first designs with stupid amounts of white space and not taking advantage of the desktops greater screen real estate and precise mouse movements, but AI first websites will get so popular that browsing sites manually will look like trying to use a text only browser in the JavaScript world.
Easy for me to do a depressing take, but hopefully the bitter lesson of AI will help this particular projected future not come to pass because the AI will get smart enough that it will embed a browser right there in line and just render the window for the user, or it will otherwise gets good enough at screen scraping and UI automation that it can just use an existing browser, just like a human, the sites won't be dumbed down even further for AI consumption.
to add to that, it kills the concept of whether the host is human (and not just another soulless megacorp harvesting your data in their walled garden)
https://news.ycombinator.com/item?id=46969751 remark that they're taking down their self-hosted projects citing costs associated with AI scraping.
at best we have walled garden content; and when those are scraped (either by the host or by more sophisticated bots) those walled gardens will hopefully rot under an inability to drive advertisement revenue.
I agree, I think we're at the edge of a paradigmatic shift away from humans navigating TCP-IP itself. What that looks like, I don't know, but given trends (like dynamic pricing, human-futures marketing, surveillance, and consolidation of computing under mega-companies) I can imagine: local beacons screaming AI advertisement components across a geospatial sneakernet. Auditorium-based ticketed podcasting and AR/VR/meatspace events. Thoughtful hackers reminiscing of better times simulating them in web-assembly driven first-person POV "sites" and a rolling set of encryption keys for read-access (just send them BTC)
without an ecosystem for humans to contribute meaningfully to a feedback loop that allows for free group assembly around like interests, monetary growth for hosts and other participants, and some degree of presence / searchability / permanence, the current text-only web page paradigm is doomed.
> AI first websites will get so popular that browsing sites manually will look like trying to use a text only browser in the JavaScript world.
That might be great for accessibility, though.
> In a world where we can type anything into a text box and get the information back instantly we are circumventing the need to visit websites altogether.
This is purely anecdotal, but the only people in my extended circle making this transition (to any extent) are the technically savvy; everyone else is slowly realizing how awful AI tools and "AI-first experiences" can be and are actively trying to avoid them.
I've noticed this bimodal distribution of perception too, and my hypothesis is that's it's hugely driven by the difference of "who is in the driver's seat".
Your tech-savvy AI early adopters are discerning between tools, the deployments and environments, and are willing and able to change things to extract the highest output from current capabilities. For instance, re-architecting a codebase to make it easier for agents to contribute to it.
The rest are having AI hypeware shoved upon them, often as a cost cutting measure, and lack the agency to influence outcomes. When agents misbehave, they only have the option to "Press 0 to speak with a Human" and hope that works.
I suspect this is a big factor in the divide we're seeing, and might result in your median adult being ambushed by recent gains in capabilities.
AI in apps is garbage. Cheap, low quality models and inflexible interaction patterns.
AI agents using frontier models, configured nicely, that interact with programs that have APIs are pure gold.
I think the more realistic direction is exposing API / MCP-style interfaces for agents to interact with a product’s functionality, rather than shipping UI components that an AI client would render.
The "AI renders your components inside chat" idea feels very similar to Facebook’s old canvas apps. That model disappeared for good reasons: abuse, security, and loss of platform control.
It seems far more likely that AI platforms will provide their own interaction primitives (forms, pickers, confirmations, etc.) and simply call third-party tools behind the scenes. That lets the platform retain control over UX and safety, and avoids the risks of embedding arbitrary third-party UI.
I've noticed a change in technical blog articles in the past 2 years. Why do most contain phrases like "everything changes", "not behind (yet)" etc.?
If you have a valid point to make, you don't need to force FOMO on the reader.
Theres probably a middle ground where we allow an app to augment / enhance itself using deterministic behaviour but a user based soft request. If I as a user can ask for a feature, and have it work just for me, thats pretty cool.
Instead of progressive enhancement it can be progressive evolution.
When I use ChatGPT to do research, I expect it to justify itself by quoting from web pages and linking to those web pages. (I gave it explicit instructions to quote things, but unfortunately it will only do short quotes.)
This might be an extended web search, but it's still a web search. The documents need to exist. Maybe a lot of the surrounding boilerplate disappears, though?
> In a world where we can type anything into a text box and get the information back instantly we are circumventing the need to visit websites altogether
But without visits many websites will disappear. So where will AI get its information from in the future?
I do like that idea of allowing flexible component rendering, especially if you're building a your own app with chat UI. The one problem would be like always standardisation, will chatUIs need to be like browser and follow standards? Or do they need to render JS, CSS, HTML as a component? What freedom do chatUIs allow the components?
Even so, ask, what's the end goal of the user? Does it even make sense to worry about UI if we're think autonomous agents that sole goal is to accomplish something defined by the user?
I was a bit confused when I went to the json-render github repo, because there were no screenshots of what it looked like.
I found only two videos about it on YouTube. This is the better of the two, and illustrates the output: https://www.youtube.com/watch?v=vndn2vmSIbw
I don't know whether that video uses Kumo (UI library also from Vercel).
> Imagine your own product. You likely won’t expect users 5 years from today to navigate 5 pages deep, apply filters and sort data just to try to derive the answer themselves… will you?
Yes, I do, because outside of tech circles, AI adoption is not as pervasive/ universal as the loudest voices keep suggesting. I’m not dismissing its value, but gentle reminder to spend more time around nontechnical people, listen to how they feel about the tech, and pay attention to how they engage with it when offered.
I like the direction this is taking, and I think the more radical and more coherent endgame is data ownership returning to us and AI's replacing website brands all together as the AI generates any UI we could ever want.
Today most websites offer interfaces for accessing the data that they hold hostage. However they are holding our data. We would prefer to have it back if we could, and AI will enable that. Our paradigm will shift this way eventually I hope.
Though it's true companies dont have an incentive to hand over data custody anytime soon, AI companies have already found a way around this in little ways and its a matter of time until it goes to where all the data we could ever want is within the AI cloud and our next task is simply taking back ownership of that (where local inference takes us).
Sadly though maybe the AI companies and our leaders-acting-as-overlords have already guessed this end game and have started to push these "subscription to laptops" and "own nothing" models with shortage of hardware and rising costs of things like Mac mini's indicating that could be a reality we experience.
So can I use this to make my pages completely unusable by AI agents, then?
> If AI wants to own the internet
As if. It's corporations that want to own _your eyeballs_.
It also kills any incentive to publish anything.
It completely removes the user from having any agency over the sources of information they are presented with.
> My hope is that you will embrace the change
Some people get so caught up with the technology they seem to forget they're charging straight into the most boring dystopia imaginable.
> You’re not behind (yet) but in today’s world
Yes, yes. We will all be left behind unless we [PLACEHOLDER]. Sounds very convincing.
Would you rather visit a website full of ads that doesn't answer your question and is written for SEO or get the answer instantly without any ads?
Eventually, after killing several websites by depriving them of revenue, ChatGPT will enshittify like everything else and starting adding ads.
There isn't even a question about that. Just think of Google for example.
Why Google's SERP has ads but Gemini does not? There isn't even a "they are making money with the data" argument here, because Google already has all the query data it could ever want. They just haven't added ads yet.
Eventually, Gemini will look like a SERP with 5 ad results, if it doesn't go to the Google graveyard like everything else.