The application: a question of accessibility and experience

Ah, our first real chapter! Here we start with the "simplest" part to grasp: what you see! The interface that lets you interact with your AI.

It may sound trivial, but it is in fact essential to a project succeeding. Very often, the way an interface hands information back to you, and how you feel using the product, determine whether users adopt it. You can have the best AI model in the world under the hood; if the interface is frustrating, nobody will use it.

The application as the real differentiator

Take ChatGPT: a simple text field (a chat) able to converse with you, write, or generate images. It was brilliant, easy to pick up, hugely intuitive!

Remember how traditional chatbots worked before that? Rigid pre-configured scripts that led you into a frustrating dead end 9 times out of 10 ("I did not understand your request, would you like to speak to an advisor?").

But be careful: chat is not the only interface, nor always the best one!

Look at the developer world. Why did tools like Cursor or GitHub Copilot take off against plain ChatGPT? Because they do not force developers to copy and paste constantly between their browser and their code editor. The AI is embedded directly where they work, inside their code.

Likewise in everyday life: would you rather ask a chatbot 15 questions to book a train ticket, or click 3 buttons in a well-designed form? The answer is obvious. AI should make life simpler, not add pointless conversational steps.

Always start from the experience you want to deliver

It is obvious when you do Product Management, but before asking how to do something (which model to choose, which infrastructure to set up), you have to ask the key question:

It sounds simple, but in 80% of AI projects today we do the opposite: we buy AI, then look for somewhere to put it.

The golden rules of good AI UX

Since AI is not software like any other (it hesitates, it takes time to answer, it can be wrong), the application has to handle particular constraints:

• 01Managing time (latency). An AI model takes several seconds to think. The interface has to show that the AI is working. That is why you see text appear word by word (streaming, the typewriter effect): it creates a feeling of immediacy and holds the user attention.
• 02Transparency and the right to be wrong. AI can hallucinate. So the application must make verification easy: show sources, offer a "Regenerate" or "Edit prompt" button.
• 03User feedback. The little thumbs up and thumbs down 👍 👎 are not there for decoration! They let the application layer collect data on answer quality, to improve the experience later.

The big trend: interfaces generated on the fly (GenUI)

Until now we had two very separate worlds:

• 01Classic applications (GUI). Fixed screens designed by designers. You click on buttons, menus, cards. It is fast, but rigid.
• 02The chatbot (NLI). A blank page with a text field. Infinitely flexible, but reading 500-word blocks of text is tedious and not always effective.

So why choose? This is where the current revolution in application design arrives: the interface generated on the fly (or generative UI).

What does that mean in practice?

Instead of answering only with raw text or Markdown, the AI generates the graphical interface you need, at the precise moment you need it.

• ▸Asking the AI to organise your trip? Rather than listing 10 hotels as text, the interface brings up an interactive map with a slider to filter your budget, and buttons to confirm your choice.
• ▸Asking it to compare two quotes? It does not write you a novel: it generates a dynamic comparison table on the fly where you can tick options or change values live.
• ▸Want to analyse your sales data? The application builds a bespoke dashboard with interactive charts, created solely to answer your question.

The XXL example: Netflix and the GenPage project

If you think generative UI is reserved for engineer demos on Twitter, think again: the giants are already at it.

In a fascinating technical article, Netflix revealed how they are testing AI-generated interfaces. The principle? Your viewing history and the time of day are passed as a prompt to a generative AI model. Instead of drawing from rigid menus, the model literally draws your app home page in real time. It adapts the layout of the blocks, the type of information shown, and even the visual arrangement, specifically for you.

Tomorrow, you and your neighbour will not merely get different recommendations on Netflix: you will not have the same application.

Why is this the future of the AI application?

Because it solves the biggest usability problem of AI: input friction.

Typing a three-line text prompt to fix one detail is tedious. Dragging a button, clicking a choice or moving a slider on an interface the AI created on the fly is fluid.

In this approach, the application is no longer a fixed mould coded once and for all: it is a chameleon that draws the graphical components based on the user intent. And for user experience, that is a genuine leap forward.

In short

The application is the bridge between the complex magic of AI models and the daily reality of your users.

Now that we have seen the visible tip of the iceberg (the interface), let us look at what happens just below when you press "Enter"… See you in chapter 3, on the software layer and orchestration!