From chatbots to generative interfaces

Topic

Generative Interfaces

Kind

Perspective

Author

Anastasia Gritsenko

Read time

4 min read

/

/

From chatbots to generative interfaces

/

/

From chatbots to generative interfaces

The chatbot may be one of the least imaginative interfaces we could have invented for artificial intelligence.

We gave software the ability to understand natural language, reason about information and generate content.

Then we put it inside a rectangle in the bottom-right corner of a website.

There is a better interface waiting to emerge.

We call it generative UI.

The chatbot was an important first step

Chat made AI immediately understandable.

People already know how to have a conversation.

There is almost no learning curve.

Ask something.

Receive something.

That works extremely well for many situations.

But websites and applications contain much more than text.

They contain products.

Tables.

Charts.

Forms.

Images.

Filters.

Comparisons.

Transactions.

Workflows.

Navigation.

Why should all of that collapse into a stream of messages?

Intent can change the interface itself

Imagine visiting an automotive website.

Instead of navigating through Cars โ†’ Used โ†’ SUV โ†’ Electric โ†’ Price โ†’ Location, you simply say:

I need an electric family SUV below โ‚ฌ45,000 that can comfortably handle a weekly 400 km commute.

A chatbot could answer with text.

A generative interface could transform the page.

The navigation disappears.

Three relevant vehicles appear.

Range and charging time become prominent.

A comparison table is generated.

Financing options appear underneath.

The nearest locations are shown.

The website has interpreted intent and assembled the interface required to fulfil it.

That is a fundamentally different interaction model.

Conversation becomes an input

The important shift is conceptual.

Chat is not necessarily the interface.

Conversation can be an input into the interface.

A user expresses intent in natural language.

The system determines which data, components and actions are relevant.

The interface then adapts.

Sometimes the correct output is text.

Sometimes it is a table.

Sometimes it is a form.

Sometimes it is a chart.

Sometimes it is an entirely different arrangement of the page.

Websites currently make users translate intent

Suppose you visit an insurance website because you want to know:

Can I insure my 19-year-old daughter on our car while she studies abroad?

The website asks you to translate that intent into its information architecture.

Which product?

Which section?

Which FAQ?

Which search term?

Which policy document?

Generative UI reverses the relationship.

The website translates the user's intent into its own structure.

That is a much more natural model.

This is bigger than personalisation

Traditional personalisation usually chooses between predefined experiences.

Segment A sees banner A.

Segment B sees banner B.

Returning customers see different recommendations.

Generative interfaces can go further.

The system can determine which combination of existing components best serves the current intent.

That could mean changing:

  • hierarchy
  • navigation
  • content
  • controls
  • recommendations
  • forms
  • comparisons
  • calls to action

The interface becomes contextual rather than fixed.

The design system becomes critical

This does not mean allowing an LLM to invent arbitrary HTML and JavaScript.

That would create problems with accessibility, brand consistency, security, performance and usability.

A safer approach is constrained generation.

Designers and engineers define an approved component system.

For example:

  • ProductCard
  • ComparisonTable
  • PriceCalculator
  • LocationMap
  • ContactForm
  • SpecificationPanel
  • CheckoutAction

The intelligence layer can select and arrange components.

It cannot invent whatever interface it wants.

AI controls composition.

The design system controls expression.

The DOM becomes an output

This introduces a different relationship between AI and front-end engineering.

Today, developers largely determine page structure in advance.

Tomorrow, applications may increasingly determine structure at runtime based on user intent.

The system interprets a request.

Retrieves relevant data.

Determines an appropriate interface.

Maps that interface to approved components.

Renders it.

The page becomes partially generated rather than simply retrieved.

Where IntentCX comes in

IntentCX is our exploration of this idea.

The goal is straightforward:

Turn existing digital experiences into interfaces that can respond to intent.

Rather than replacing a website with a chatbot, IntentCX adds an intelligence layer capable of changing how the existing interface behaves.

The website remains a website.

The design system remains controlled.

The underlying APIs and content remain available.

But the experience can reorganise itself around what the user is trying to accomplish.

E-commerce is an obvious starting point

Consider:

I need everything required to repaint a 20mยฒ bedroom this weekend. The walls are currently dark blue and I want them off-white.

A traditional e-commerce site returns search results for paint.

A generative interface could understand the complete job.

It could assemble:

  • primer
  • paint quantity
  • rollers
  • brushes
  • masking tape
  • floor protection

It could calculate quantities.

It could check inventory.

It could organise products by the sequence in which they will be used.

It could create one basket.

That is closer to having a knowledgeable employee standing beside you than using search.

Enterprise software may change even more

The same principle applies internally.

An employee asks:

Prepare me for tomorrow's meeting with Acme.

The interface generates:

Company overview.

Revenue history.

Open opportunities.

Recent correspondence.

Outstanding issues.

Meeting attendees.

Suggested discussion points.

Relevant documents.

The application is no longer a collection of fixed screens.

It becomes an interface assembled around the task.

What happens to UX?

Design becomes more important.

Someone still needs to define what the system is allowed to generate.

Which components exist?

Which combinations make sense?

Which information deserves prominence?

Which actions require confirmation?

What happens when intent is ambiguous?

How does the interface explain why it changed?

How does it remain accessible?

Generative UI moves design from creating every screen towards creating the rules and components from which screens can be assembled.

Beyond the chatbot

Chatbots will remain useful.

But they are unlikely to be the final interface for AI.

The more interesting future is software that listens to what you want and changes itself accordingly.

Menus become optional.

Search becomes conversational.

Pages become contextual.

Interfaces become generated.

For thirty years, users learned how websites worked.

The next generation of websites may learn how users work.

Avatar-1
Avatar-2
Avatar-4
Avatar-4
Avatar-3

Fresh insights, straight to your inbox.

Engineering breakdowns, sovereign-AI perspectives and practical guides from the people building enterprise AI. One email when there is something worth your time.