Criterio — 02

When the interface stops being a screen

The interface hasn't disappeared. It has left the screen.

“The interface hasn't disappeared. It has left the screen.”

Designing products with AI is no longer only about organizing buttons and journeys. It also means defining what the system understands, what it can decide and how far it can act.

For years, designing a digital product meant organizing information, buttons, menus and journeys so a person could complete a task. The logic was fairly clear: the user decided and the system executed.

Artificial intelligence changes this relationship. Now the user can express a goal and let the product interpret, propose or even act on their behalf.

We no longer design only what a person sees. We also design the behavior of the system.

Not everything needs a chat

One of the most frequent mistakes when adding AI is turning any experience into a conversation. But a conversational interface is not always the best solution.

If the user's intent is clear and the information is structured, a form is still faster and more precise. If the need is ambiguous, requires comparing options or exploring possibilities, a conversation can add more value.

When the task is repetitive and controllable, it can be delegated to an agent. If it involves economic, legal or hard-to-reverse consequences, the system must hand control back to the person.

NeedSuitable interactionReason
Change an addressFormIntent and data are clear
Choose a mortgageConversation and comparisonThere are multiple criteria and consequences
Prepare the weekly sales reportAgent with reviewIt's a repetitive, delegable task
Make a transferStructured flow and human confirmationThere is real financial risk

The right question is not: “Where do we put the chat?” It is: “What kind of interaction does this decision actually need?”

Designing the system's limits

When a product incorporates AI, the experience no longer ends when you send an instruction. You have to define what happens next. Before letting the system act, any team should answer five questions:

1. What can it do on its own?

An agent can gather data, organize information or prepare a draft. But it perhaps shouldn't send it, publish it or approve it without supervision. Autonomy should depend on the risk, not on how spectacular the technology looks.

2. What must it explain before acting?

If an AI recommends a mortgage, rejects an application or changes a forecast, the user needs to understand the criteria used. An answer without context can seem fast. It can also be completely opaque.

3. When does it need permission?

Not all actions have the same impact. Preparing a report is not the same as sharing it with the whole company. Calculating a transfer is not the same as executing it. Human permission must appear at the right moment: not too early, blocking the experience, nor too late, when the damage is already done.

4. How is an action corrected?

Products with AI need clear mechanisms to edit, stop, repeat or undo. If the system can act, the user must be able to regain control. Without reversibility, automation stops being a convenience and becomes a risk.

5. Who answers when something fails?

A good experience must indicate what happened, what data was used and how the problem can be escalated. “The AI did it” is not a strategy of accountability. It's an excuse with better branding.

Trust is designed too

A clean interface doesn't make up for an unpredictable system. To trust a product with AI, people need to know:

This transparency is not an extra layer of communication. It is part of the product.

Experience is no longer measured only by ease of use. It must also be assessed by the system's ability to act in an understandable, controllable and coherent way.

The advantage won't be only in the model

Models will be increasingly powerful and accessible. The difference will be in how each company turns them into genuinely useful products. That requires connecting four dimensions:

A powerful AI, poorly integrated, doesn't necessarily improve an experience. It can accelerate wrong decisions, increase complexity and multiply errors.

Innovation isn't about automating everything. It's about knowing what to automate, for whom, under what conditions and with what guarantees.

From the interface to the architecture of decisions

At EIDOS we tackle this challenge before designing a single screen. Through FORMA, we connect strategy, technology and design to define:

Because the future of products with AI won't depend only on what the technology is capable of doing. It will depend on our ability to turn it into better decisions, understandable experiences and real outcomes.

Better decisions. Better outcomes.

Reading that inspired this reflection: The Interface Has Left the Building.

Back to Criterio