Speaker A
Hey, everybody, thank you very much for joining us. My name is Alan. I'm joined by several other peers.
Explore generative UI innovations with MCP apps, enabling interactive AI-driven interfaces across platforms and agents.
Key Takeaways
- Text-only interfaces are insufficient; generative UI offers richer, interactive experiences.
- MCP apps standardize UI transmission and interaction protocols for AI agents.
- AI agents can dynamically create bespoke UIs tailored to user needs in real time.
- The future of web interaction involves modular, agent-driven UI components rather than monolithic apps.
- There is a huge market opportunity as billions use chat-based AI platforms supporting MCP apps.
What the video covers
- Introduction to generative UI and the limitations of text-only interfaces.
- Explanation of MCP apps as an extension to MCP for creating interactive, graphical UIs.
- Demonstration of replacing textual responses with branded, interactive UI components.
- Discussion on bespoke UIs generated on the fly by AI agents for enhanced user experience.
- Comparison of traditional complex web UIs versus agent-driven modular UI components.
- Overview of the spectrum of UI from predefined apps to fully generative UIs.
- MCP apps' interoperability and support by major AI hosts like OpenAI, Claude, and Gemini.
- The massive user base for chat-based apps and the opportunity for developers.
- Encouragement to rethink product design as interconnected web applications powered by AI.
- Introduction of related protocols and the vision for agent-driven, adaptive user interfaces.
Chapters
- 00:00Introduction and Overview of Generative UI
- 01:55Demonstration of MCP Apps with Funnel Status Example
- 03:30Interactive and Bespoke UI Generation by AI Agents
- 05:10Breaking Down Complex Web UIs for Agent Use
- 06:45MCP Apps Agnosticism and Interoperability
- 08:09Market Adoption and Scale of MCP Apps
- 09:50Challenges and Solutions in Agent UI Development
- 11:13Declarative UI and Developer Tools
- 15:10Generative UI Demos and Production Support
- 19:00Intent-Driven Interfaces and Future Directions
Full Transcript — Download SRT & Markdown
Speaker A
This is a partnership conversation with a lot of different players talking about generative UI.
Speaker A
I'm very excited to be here. We have a lot of good content. We're going to try and make it through all of these slides.
Speaker A
We're going to talk about my and many of your frustrations with a text-only interface, and how we believe that all of the interfaces will soon have some sort of AI behind them, and you can do a whole lot more with it.
Speaker A
But I'm actually not going to start this talk. I'm going to hand it over to Ido to come here and talk about MCP apps to get us started.
Speaker A
Thanks, Ben. So hi everyone. Thank you, Alan. Sorry. So I'm Ira Solomon. I am the creator of MCP and Agent Craft and co-creator and maintainer of MCP apps.
Speaker A
Before we get started, I just wanted to shout out to the amazing people that made MCP happen.
Speaker A
So these are co-creators. We have Liad and Olivier and Anton and Nick and a bunch of other great people.
Speaker A
So as Alan hinted at, text isn't enough. So what is? How do we transition from text into rich graphical user interfaces?
Speaker A
That's exactly why MCP apps was created. It's the official extension to MCP. It's how we do interactive apps over MCP, standardizing transmission of UI to clients and the communication protocol between clients and UI.
Speaker A
Let's see how it works just to get an impression. So, for example, let's say that I have a product and I want to understand the status of my funnel.
Speaker A
So I ask my agent and in the old world it would reach out to my analytic provider, like PostHog, and we'd get back this textual response, this wall of text, which is nice and factually correct, but doesn't really help me understand
Speaker A
what the hell is going on. Luckily, because both Claude in this instance and PostHog support MCP apps,
Speaker A
instead of that, I can just say, show me. And now instead of getting that—sorry, instead of getting those—
Speaker A
sorry, clickers messed up. OK, so because they both support it, then we can say show me.
Speaker A
And now instead of text I can get this interactive UI component created by PostHog with its own branding and user experience, and I can see what's going on with my funnel at a glance.
Speaker A
Now, we don't want to be constrained only to things that are predefined in some application.
Speaker A
We do want to let the agent run free and create bespoke UIs for us.
Speaker A
For example, let's say that we want to understand what a funnel is. So what we can do is just tell the agent, explain the funnel to me.
Speaker A
And now instead of writing something that I need to read and understand, it could just generate something on the fly.
Speaker A
This nice UI explaining to me what a funnel is and how it works. Now, this isn't just presentational.
Speaker A
As we said, apps also support interactive interactivity. So actually embedded into this UI component is interactivity.
Speaker A
So let's say that I want to understand more about a particular phase in the funnel.
Speaker A
I don't need to prompt it anymore. I don't need to figure out what I need to say to get what I want.
Speaker A
The model already embedded a behavior that when I click on, show me one of these steps, it will actually send a prompt to the agent to drive forward the flow and give me that explanation that I need.
Speaker A
So we saw technically what MCP apps is capable of, but if we zoom out for a second, it's not just nicer chat.
Speaker A
It's actually a fundamentally new way to interact with the web. If we look at the old web, for example, even doing something simple like preparing for a special occasion would require me to go to 10 different websites and understand how do I book something, how
Speaker A
do I order some flowers, how do I do a bunch of stuff. And if you look at these apps or these websites, they're really complex.
Speaker A
And the reason that they're so complex is that they're meant for us humans. And you don't know as a website, what do you mean is actually trying to do.
Speaker A
So you just push everything that the human might do into one screen, and you need to simplify it or build it in some complex way
Speaker A
so you'll be able to consume it reasonably. But if you take these complex UIs and break them apart, you suddenly realize that agents aren't like humans.
Speaker A
Agents know exactly what you want to do, so they can take just what they need.
Speaker A
Just a map, just a particular checkout button, whatever, and present it to me directly.
Speaker A
So in this example, all we actually needed to get this flow going is the map to see where it is and maybe the hotel and what that flow would actually look like in an agent with MCP apps is
Speaker A
that it would surface the date for me, just the UI of that date. It would surface just the shopping aspect,
Speaker A
so just the flowers that I want to buy. Same goes for the hotel itself, but I would say it's not just limited to predefined applications.
Speaker A
We also want the agent to be able to run wild and create bespoke UIs just for what I need to really complete that end-to-end flow.
Speaker A
So if we look at what we just saw, there's kind of a spectrum of UI. Like we have the predefined UI, we start with PostHog.
Speaker A
It's just an app. It's just like a website that you've built until today. The second step is going towards the declarative UI, which is more of you describe basically what type of components you want, and the host builds it the way that it sees fit.
Speaker A
And the third one is what we saw with Claude. Generative UI is really having the model create content from nothing.
Speaker A
Just stream that HTML into the canvas, into the app, and display it. The beautiful thing here is that MCP apps doesn't care.
Speaker A
MCP apps is agnostic to what's being rendered. That's actually what makes it so powerful in interoperability with other protocols.
Speaker A
And as you're going to see here soon, you have a UI and UI that are practically built into it because they can be rendered inside the MCP app and conversely, the MCP app can be used as a component
Speaker A
that these frameworks, these protocols can render. So you can have really rich experiences in the chat.
Speaker A
Now MCP apps are everywhere. I don't know if you're aware of it, but basically every major host supports it.
Speaker A
So we have OpenAI and Claude and Gemini and a bunch of others just looking at ChatGPT and their apps.
Speaker A
They have over 500 apps from huge companies like Walmart and Salesforce and you name it.
Speaker A
So this isn't something that this is just a concept. It's much more than that.
Speaker A
It's just the new app distribution surface. If you look at, again, ChatGPT, for example, they have one billion weekly active users, which is 12% of the world's population.
Speaker A
If you add Gemini and Claude and other hosts, the number of users that are using chats is much bigger.
Speaker A
That's over 160 times the number of users that the Apple App Store had when they launched.
Speaker A
So as you can see, it's a huge total addressable market. And the question that you need to ask yourself now isn't why you need to develop an app.
Speaker A
Why do you need to get your product ready for the gigantic web? It's what happens when users stop being the people that see my website.
Speaker A
It suddenly becomes agents. So I hope you look at this thing as a unique opportunity.
Speaker A
Just to take a step back, look at the products that we're building, take them apart, and build them back.
Speaker A
Not as a monolithic app, but actually as part of a web of applications connected together by the models and with MCP apps and other protocols like E2E and Agwe, you can really build your product once and run it everywhere.
Speaker A
Thank you. Thanks, Ido. Atoui, before I get into that, my name is Alan Blunt.
Speaker A
I was a full stack web application developer for 20 years and then did some machine learning stuff in Google Cloud, and now I'm doing agent building platform.
Speaker A
And a lot of the protocols that you have heard of Atoui is one of that list.
Speaker A
We are actually a hybrid team collecting a lot of people that have contributed for over two years towards this initiative is what you are seeing now.
Speaker A
This team is trying to solve basically these three sets of problems. When I create an agent on my application and put it on
Speaker A
But if I have a remote agent that is part of my system or a mesh of agents, how can they communicate.
Speaker A
How do I do true generative UI. And how do I get that to users safely, wherever they are.
Speaker A
It's sort of a complicated action space, but this is what we're trying to solve for with atoui.
Speaker A
Let's start with remote agents, and let's start with a diagram. This is just one configuration.
Speaker A
There are many configurations, but it sort of sets the stage. So I've got my design system, my front end application which is where my users are clicking and doing and typing.
Speaker A
That front end is connected to some backend, some backend server. That is my APIs, my agents, my whatever.
Speaker A
That's a very normal pattern that we all have. But maybe I have these agents that are not in my system, that are in some remote system.
Speaker A
That is a particularly hard part of this equation, especially when you get into users and ACLs and all of the other parts and pieces of the system that need to be secure.
Speaker A
So we really needed a way to do message passing across these entities in a way that we could secure to treat UI as if it was just data.
Speaker A
So we don't need all of that. Let's start with a simpler scenario. If you have a front end and you have a back end, and maybe that back end has some way of generating a UI, an UI message,
Speaker A
you could just ship that across. It doesn't matter if you have subagents that are local subagents you're in the happy simple scenario path totally works.
Speaker A
Ooh, I did want to point out that transport is totally agnostic. You don't need a two way you can do it over rest.
Speaker A
You can do it over web sockets. You can use aighewi. You can use MCP for your front end.
Speaker A
It doesn't matter. But sometimes you may collaborate with other remote agents. Those remote agents are the ones actually generating that E2E message, and you're just passing it across through the orchestrator because it's just a normal message payload.
Speaker A
And we'll get into that in a second. It's very easy to simply pass across.
Speaker A
One other permutation is the remote agent may be doesn't know how to generate an E2E message, but your orchestrator does.
Speaker A
You can manipulate or mutate that message as it comes across if you need to.
Speaker A
These are just a few of the patterns. There's many, many, many others, but I wanted to set the context.
Speaker A
Now the generative UI part, this is not using tool calling. This is just structured output.
Speaker A
We chose json-ld because many LLMs, most LLMs are pretty good at JSON. There are other types of declarative UI formats that we could use.
Speaker A
We picked a thing that worked for us today. It could certainly evolve if we needed to.
Speaker A
The LLM assembles these components together. We'll talk more about the components in a second.
Speaker A
These symbols, these components together, maybe one component, maybe multiple components into a widget or a surface is what we call it, some kind of a layout.
Speaker A
And then it hydrates those variables in that surface or populates the variables. These are streaming chunks of messages.
Speaker A
So it's a pretty good user experience. And again, very flexible. And one of the use cases I'm very excited about is cacheable configurable UIs. So your UI is already configured full UI application, but it can be mutated whenever you
Speaker A
need to mutate it by an agent. That means you can customize your whole application just by asking your agent to change a button or I never need this thing before.
Speaker A
And it gives you the flexibility to control that UI. So we start with the basic catalogs that we used to call them standard.
Speaker A
Now we renamed it to basic to try and make this more clear. It has a handful of components in it.
Speaker A
You can use them. They help you get started quickly, but we expect most of you already have a front end.
Speaker A
You already have your own button component. We want you to be able to use your button component and assemble it with all of the rest of your components to make these widgets.
Speaker A
And you put the widgets together. They're interactive and you don't have to use just flat components.
Speaker A
You can use fancy components that are dynamic. You hydrate them with whatever data sets are available to you.
Speaker A
You define a custom catalog of which are the components that you're allowed to use.
Speaker A
And that kind of becomes a contract between the front end and the back end.
Speaker A
Whatever the client advertises, the back end can use. And I think I'm repeating myself, but I'm just going to say it again.
Speaker A
You already have a design system. You already have a front end. You don't need to change that.
Speaker A
I want you to be able to make your front end. Maybe it's material design, maybe it's whatever accessible to your agents and remote agents.
Speaker A
All right. And so we are largely talking about the web. I often interact through the web, my laptop.
Speaker A
But let's face it, I'm in meetings all day and I'm running around and I'm often on my phone, and then occasionally I'm on a tablet.
Speaker A
And then every once in a while I have something on a TV and then I'm on my watch.
Speaker A
And like, these are all different types of devices. And while I love the web, and I have built most of my career trying to build things for the web, there are lots of other interfaces out there.
Speaker A
And we want to build a solution that works across all of them. There are some very, very interesting developments happening on all of these surfaces, but I don't have dates and times yet.
Speaker A
But keep watching. There's a lot of really interesting stuff happening across Google and also customers.
Speaker A
I've talked to two or three customers that have bespoke rendering pipelines and bespoke interfaces, and they're very excited about a standard for declarative UI.
Speaker A
This is not new. There have been standards for declarative UI before, but this is a great moment to realize that.
Speaker A
So why do I care so much about the surface. Security is one of the aspects we're all familiar with data exfiltration and rendering an image that allows an agent to accidentally leak data that's a lot easier to protect against if you're in control of what
Speaker A
the image component is, and if you have structured out or decoupled the values for that image, you have layers of defense similarly.
Speaker A
Hidden forms inside of a click target is another attack vector that you want to protect against.
Speaker A
And the reason, sorry, the way that you can protect against this is by being in control of your component library system.
Speaker A
Or we've seen a lot of different supply chain attacks. One of the ways that you can protect against this is by being in control of all of the front end components that you have.
Speaker A
And again, this is on you to control. But this allows you layers of defense.
Speaker A
And that's really what the security concerns are. Are defense in depth in UI. There's a contract between the agent or all of the agents in the background and the client.
Speaker A
The client is in charge of what components are available. There's lots and lots of other security concerns, but know that was one of the main designing principles in this.
Speaker A
So what you get is a set of core libraries that handle a lot of the plumbing and the message passing for you.
Speaker A
Some renderers that we maintain, you're in charge of your client catalog and pass that in, and your agent sends the message across and it shows up for your user.
Speaker A
Where are we at today. Do we have any transport, any catalog of components. We do have some tooling that repairs generations if they mess up a little bit on the fly.
Speaker A
We have a handful of libraries that are renderers for Web Components or lit. Angular React.
Speaker A
Flutter these are contributed to by the main owners of these libraries. And then we have an ecosystem of other renderers that have happened around that people are contributing to us.
Speaker A
Today, a-to-I is launched in Gemini, enterprise and Opal and many other internal tools. There's lots and lots of teams exploring it.
Speaker A
These demos don't really show the full power. I'm still pretty excited to get just an input, just a button.
Speaker A
I just want a little bit and I get excited, but there's so much more that is coming, as you'll see a little bit later.
Speaker A
We've got some people already exploring generative UI and coming up with much more exciting demos than this, but I personally am still very excited to see this inside of our products.
Speaker A
Showing better information, coming back, showing a few better affordances, and rendering fancier components or widgets whenever we need them.
Speaker A
We also have a few experiments going on, some of which are public, and we'll make a few more about mixing and matching between MCP, apps and atooi and some other dynamically generated UI widgets.
Speaker A
Where else are we going tomorrow. More integrations and more products. Again, I'm not. I cannot leak any of the names or things, but there are things coming that I'm very excited about.
Speaker A
Stability is a big concern for me. I really want this to get to a 1.0 release this year so that we can have long term support and feel very consistent and stable on it.
Speaker A
I do think that the core infrastructure is now quite stable. We're going to add some new features with the release, and then we're probably going to try for av1 point 1.0, hopefully.
Speaker A
Q3 and we're going to make it easier to use. It has largely been get all the parts and pieces right and make it work.
Speaker A
But we have not yet made it super, super easy for devs. We made some big strides towards that and the 0.9 release, but we certainly have more Polish to do for the developer experience.
Speaker A
And speaking of making things easier for developers, actually I would like to invite up a Ty to come explain.
Speaker A
Aigoo, aigoo to you. Hey everybody. Thank you. Alan can everybody hear me. All right.
Speaker A
Great all right, I'm Ty. I'm co-founder and CEO of copycat. For those who don't know us, copycat makes developer infrastructure for building AI Copilot, co-pilot, which essentially means user interactive agents as opposed to fully autonomous agents.
Speaker A
We make the layers that connect the agentic world with the user facing application world, and at this point powers tens of millions of agent user interactions per week in production from companies like Cisco, Docusign, function health, Deutsche Telekom and in fact, the majority
Speaker A
of the global 50 as well as the Fortune 500 are today Copilot users. We're also the company behind the agwi protocol.
Speaker A
If you might heard of it. It's a protocol that emerged from our partnership with first chain and then kuii, and has since been adopted by essentially all the top leaders in the space, including Google, which has been incredible to collaborate
Speaker A
with in Amazon, Microsoft, Oracle, and essentially all the top AI framework startups. So master, pedantic, ag2 and so on, I'm really excited to be here today and to work in this space.
Speaker A
And the reason is we believe over the next few years all UI will be AI, which means that over the next few years, all interactions between people and technology are going to become increasingly mediated by agentic systems.
Speaker A
And obviously, this is true of complex software like HubSpot and Figma and productivity software like Cursor Cloud Code.
Speaker A
But it may even be true of your refrigerator. And, the first time I heard this, when we were approached by a large electronics manufacturer, I thought it was a little bit funny.
Speaker A
But on second thought, wouldn't it be nice if you could walk by your refrigerator and tell it to order the ingredients that are missing to make lasagna tonight.
Speaker A
So that's the world we're going to. But where are we today. Many of you audience probably know that building great user interactive agents is very difficult. And the core reason is that agents break the request response paradigm that's
Speaker A
been powering the internet for the last 30 years, right. So agents are, of course, just software.
Speaker A
But from the standpoint of software, they're a little bit like a weird bird. They have their long running, so they have to stream their work over time and support reconnections and even mid-run steerings.
Speaker A
They have to support structured and unstructured Data Exchange at the same time. So they have tool calls and state updates at the same time.
Speaker A
You have text and voice. There's really a long list of these peculiarities and difficulties which make it challenging to build great generic applications.
Speaker A
And that's where Copilot and ag come in. So Copilot provides open source developer SDKs and a self-hostable Cloud for building generic applications in a way that's optimized for the generic paradigm, as opposed to retrofitted from request response in AG, which again stands for the agent user interaction
Speaker A
protocol lets developers who build agentic frameworks. And harnesses bring their generic backends ends to any custom hygienic front end by building to a robust standard.
Speaker A
So it makes it very easy, very simple to conform to the requirements and everything just works and continues to automatically evolve with the ecosystem and to situate it in the landscape.
Speaker A
You obviously have MCP connecting agents to tools into context and to connecting agents to other genetic systems.
Speaker A
AGI is the third leg of this triangle connects your agents to user facing services.
Speaker A
Our initial focus for AGI has been the web, but as Alan mentioned, there's quickly growing list of front services supported by AGI, including Slack, WhatsApp, mobile and other services.
Speaker A
Here's an architectural overview of AGI. We actually don't have time to get into this, so you want to take a picture and ask me questions about it afterwards, I'll leave it for a moment.
Speaker A
All right. What's become very clear to us, working with all these top leaders over the past couple of years is that no single generative UI solution is good for all different use cases, and generative UI solutions really live on a spectrum that extends from more
Speaker A
control to fully open ended. And AGI supports this full spectrum, which means you can bring the right tool for the right job at any given time.
Speaker A
And now I'm going to speed through the three key pillars of the generative UI spectrum controlled, declarative and open ended generative UI.
Speaker A
So first control of UI wear developers define a collection of pre-built components which the agent can call upon when he wants to show the UI.
Speaker A
So first, let's just see it in action. So here we see a user asking to see revenue split by category, and instead of getting kind of an impenetrable long paragraph of text, you see this beautiful pie chart that is much more easy to understand from a code
Speaker A
standpoint. It's extremely straightforward. Here, for example, we have the use component hook in React where you define the component and what it's for the parameters.
Speaker A
Whoops the parameters that it requires in just a React renderer. In this case a pie chart.
Speaker A
To some extent controlled generative UI is the most boring part of this spectrum, but it does give you Pixel perfect designs and maximal determinism, which makes it a really great fit for the few most used surfaces in your applications.
Speaker A
For example, if you're an airline, you want your flight cards to be Pixel perfect and as deterministic as possible, and for that reason, we see control UI is the workhorse of the generative UI spectrum.
Speaker A
Next, we have declarative UI where UI fits in and where developers declare a component catalog of LEGO like building blocks that the agent then assembles on demand to answer any user queries.
Speaker A
So let's see that in action. All right. So here we see a user asking for a sales dashboard that includes revenue and customer churn.
Speaker A
And it includes all the information a salesperson might want to see instead of again getting a block of text.
Speaker A
You see this beautiful sales dashboard that combines cards with titles and pie charts and bar charts.
Speaker A
Each one of these components individually was predefined by the programmer, but the collection of them together as a sales dashboard was assembled by the agent on demand.
Speaker A
And to look at some code to define this, you first define the catalog definitions, what each component is and what arguments it acquires from the outside.
Speaker A
You'd need a catalog renders to render those this data in React in this case.
Speaker A
And that's it right things do the AG A2 UI handshake. That's a mouthful. That's all you need to do to get any AGI agent, which is again, essentially all of them to speak to UI to your users in your own custom application e.g.
Speaker A
middleware absorbs all the additional complexity, and with declarative UI you're giving up some Pixel perfect perfection.
Speaker A
You define the components, but every component has to be an assembly of these pre-built building blocks.
Speaker A
And you're giving up some determinism because the agent is assembling these components together, typically.
Speaker A
But what you get in return is very broad support for different user queries with a one time definition of a component catalog.
Speaker A
And this profile trade offs makes it a really great choice for the long tail of user facing services in consumer applications, and also for internal enterprise applications where the efficiency of implementation is really important.
Speaker A
And finally, at the far end of the spectrum, we have open ended UI. So in the previous two approaches, we at build time prescribe what types of UIs the agent can show with open ended generative UI, you're opening the door for the agent
Speaker A
to display anything at all. And typically that's achieved via an embedded iframe. And there are two kinds of open ended UIs that are worth calling out.
Speaker A
Of course, MCP apps that I showed us before. They're really optimized for third party tools that you want to bring into your own application.
Speaker A
And again, there's a large collection of these that are built for the Super host.
Speaker A
For ChatGPT for Cloud for cursor and so on. And then the second kind is fully open ended generative UI, where the agent actually generates fully custom HTML live inside of a secure sandbox for security purposes, but they can show really anything at all.
Speaker A
So we already saw a demo of MCP apps. So here's a demo of fully open ended generative UI where the user asks for a specialized calculator for some specific use case.
Speaker A
They have even a specific design that they want, and the agent is building it on the fly.
Speaker A
And as you can see, it's a little bit slower. It's less deterministic, but you do get at the very end and actually fully functioning component that nobody could ever think to design ahead of time.
Speaker A
And again, with AG Aggie MCP, apps and open ended generative UI are just a configuration away in your own custom full stack agents, so you can pass the URL of any MCP apps you want to use and just turn on the open generative UI flag.
Speaker A
And now any of your backend agents without any further customization in the backend can support these experiences.
Speaker A
So to sum up, we covered the three pillars of the generative UI spectrum, starting with controlled generative UI where you define pre-built components that are good for your few most used surfaces.
Speaker A
Then declarative generative UI of LEGO like building blocks that are assembled together, which are great for the long tail of UIs and for internal enterprise applications.
Speaker A
And finally, fully open ended UIs, which are really great for third party applications and for fully custom experiences.
Speaker A
If you'd like to speak about this, you can email me while we're all in town.
Speaker A
My email is atai at. Capecchi also be outside here answering questions, doing some impromptu demos.
Speaker A
And we actually have a special offering for folks at Google Cloud Next to help your engineers ship your applications to production so you can scan the QR code if you'd like.
Speaker A
And now I'm really excited to bring Nico from Akka to show us how all of this theory is put to practice.
Speaker A
Thanks sati. Hi everybody. So now I'm going to talk to you about why brands and big enterprises needs to get into January.
Speaker A
I'm Nicolas, I'm the CTO EMEA at akaaka. I'm part of WTP. Why it's important.
Speaker A
There is a structural shift. For the past 20 decades, we have built websites which were optimized around navigation.
Speaker A
We take it for granted, right. You navigate browse, Bryce. You go to static pages.
Speaker A
And from that, companies were inferring your intent, and we could personalize experiences. But now this is changing.
Speaker A
You're doing the discovery of products and services already in ChatGPT in Gemini. And when you get onto a website, you already have a clear goal.
Speaker A
And either you find the answer from this goal on the page, or you will move back to ChatGPT and Gemini to continue your searches.
Speaker A
So navigation is not enough anymore. We need solutions that can adapt. That takes us to the content economy.
Speaker A
First of all, it's built intent is the primary signal. Secondly, you need systems which are adaptive, which assemble, which are not fixed.
Speaker A
And the last one is changing the trust between the system and user. Because now we are actually capturing real intent.
Speaker A
We're responding to an intent. We're not trying to track you and infer what you want.
Speaker A
So if intent becomes the primary signals, there is a fundamental challenge for enterprises. Because human intent is rich, it's messy.
Speaker A
It's nuanced. Enterprises can't work with messy elements like that. So we need structured intent.
Speaker A
How do we do that. Well, we go back to work that was done in the 90s as the jobs to be done framework.
Speaker A
And it tells us that human intent is actually built on three elements. The functional part.
Speaker A
What are you looking for. The size, the color of the object you're looking at.
Speaker A
But there's also the part which is super important. What is the confidence level. What is the risk level.
Speaker A
What is the urgency of your need. Very importantly, the social part. How do you want to be perceived.
Speaker A
The politics at play. The pressure. So real decisions are made with intent. Object functional.
Speaker A
Emotional and social. So we're going to look at how we do that with a real world scenario with a company called synthesis.
Speaker A
It's a fraud detection platform which helps banks, fintech and e-commerce e-commerce detect fraud at payment level.
Speaker A
It's a fictional brand, but it's built on real work. I can't share the real work because it's confidential, but we rebuilt the demo with the synthesis.
Speaker A
So now I'm going to show you actually in the backend how we're capturing the intent of a future consumer.
Speaker A
So here, it's taking the intent from a new security officer in a startup, which the recent rises in cross-border payment.
Speaker A
And what we're doing here is our Gemini agent is splitting that intent into functional, personal, Lord, personal and emotional.
Speaker A
From a functional point of view, it's telling us that person is looking from a fraud detection solution, emotional point of view.
Speaker A
That person needs to feel competent in his choice and from a social element. Very importantly, we assume leadership pressure for him to make this choice.
Speaker A
This solution is creating an intent object that gives us product matching the tone of voice that we need to use to talk to that user.
Speaker A
And the experience design we need to create. So how do we go from intent to interface now.
Speaker A
Well, three elements. The interpretation layer. We've just seen it. We've taken intent. We break it down.
Speaker A
Then we pass that to a brain. Brain this is where all of the product data, the brain data is stored.
Speaker A
And it also has governance and compliance. And finally, we pass this to the generative interface which takes this content and assemble it on the fly.
Speaker A
The separation is very important because we want enterprises to be able to control this system.
Speaker A
Let's look at the brand right now. So brand brand at the core has a vector database which has all of the product data.
Speaker A
But we also have a knowledge graph here, which has all the complex enterprise context.
Speaker A
And finally, we have a layer dedicated to compliance legal compliance for from the company.
Speaker A
And the separation of content is very important again. So we give control to companies right.
Speaker A
We don't want to be an agent in a black box where everything is controlled in prompt.
Speaker A
The output of the brain drain is a validated content object that is given to the generative interface layer.
Speaker A
So it takes that intent structured content. Then it will select the components from the design system and assemble the layout and inject the content into it.
Speaker A
This is then always done within the design system constraint. So let's look at a first demo of that website.
Speaker A
So here this is the synthesis homepage. You can see standard homepage talking about the actual product.
Speaker A
The USBs. But at the top of that page we allow that visitor to enter his real question his intent.
Speaker A
And here it's someone that wants to improve fraud detection without hurting conversion. So it's giving it that intent.
Speaker A
The system we're going through all the layers the brain, brain, the interpretation layer. And we actually generating a completely new page with the right component to answer his needs.
Speaker A
So here we have generated a full solution pages with our reassurances. All of the content that is here has been rewritten by the brain, but the components is used from the design system.
Speaker A
It has, as you can see, some Roi which has been picked specifically based on that intent.
Speaker A
Some specific content about fraud detection. Pull that quote that makes sense based on the user search.
Speaker A
So let's now look at a different intent from a different user. In this case, it's someone which is looking to fix fraud spike extremely fast.
Speaker A
They've seen fraud happening and they want something that fix it in weeks, not months.
Speaker A
So the system gets that intent and will build the page. At this point, the system understood the urgency, right.
Speaker A
The social and emotional aspect of that. And it's been on the page with a decent system.
Speaker A
But here it's a much more focused response. It's not full detailed page. The pages is created to go straight to the next best action for that consumer.
Speaker A
So for that team that we work with, what it means is these type of responses.
Speaker A
They used to do it by hand create. They created PDF created by humans. It used to take them more than six hours.
Speaker A
Now, this is done in tens directly on their website. So massive acceleration. They've also clearly counted how much money they were spending creating this dedicated response.
Speaker A
And that cost obviously now is gone. And more importantly, they're telling us that we're no longer spending hours preparing responses.
Speaker A
We can focus on the strategy discussion from the start. It's really changing how this company is engaging with its leads.
Speaker A
So now content and is infrastructure for company navigation is replaced by intent. Static pages needs to become dynamic system.
Speaker A
So as enterprises for the next web experiences, you need systems that understand reason and Compose in real time.
Speaker A
And I will give it back to Alan. Thank you. It has been a privilege and a joy to work with the other presenters that you saw today, with all of the engineers that have got us here.
Speaker A
And honestly, with all of you, we are working on things that are new and we're coming up with really good foundational technologies and capabilities ourselves, but we're doing it together to build the new versions of the web.
Speaker A
I'm putting this QR code here and on this thank you slide at the end.
Speaker A
Same QR code to a form. It's an anonymous form. You can put in contact information if you want to.
Speaker A
This is what would you like from generative UI projects in 2026. We are working together at least monthly, often much more frequently.
Speaker A
We will continue to do so this year. What would you like to see from any of these projects.
Speaker A
Please tell us. We'll also be hanging out in the hallway or whatever, and you can come find us here.
Speaker A
But I do want to hear from you. We are solving our problems and the problems that we see and the problems that we hear from you, but this is an opportunity.
Speaker A
Please tell us what you want. We are all in this together. And thank you for joining me.
Topics:generative UIMCP appsAI agentsinteractive UIGoogle Cloud Techagent-driven interfacesdeclarative UIOpenAIClaudeGemini
