When the Interface Becomes the Barrier

What we are learning about intent-based UX/UI, neurodiversity, and human capability

At Adonea.ai, we have been thinking deeply about accessibility.

But recently, something changed.

We began to realize that accessibility may be the wrong place to start.

The deeper question is not simply:

How do we make an existing interface accessible to more people?

It is:

Why are we asking humans to adapt themselves to an interface in the first place?

That question has begun to reshape how we think about human-centered AI, assessment, employment, and what we are calling intent-based UX/UI.

When a test measures the interface

One of the experiences that pushed our thinking forward came from looking closely at standardized testing.

An assessment might claim to measure reasoning, knowledge, judgment, or problem-solving. But the person taking it must also navigate typography, screen density, visual organization, timing, navigation controls, information hierarchy, and numerous other elements of an interface.

For many people, those things are barely noticeable.

For others, including people with dyslexia, ADHD and other forms of neurodivergence, they can consume significant cognitive resources.

Then something troubling happens.

A person may understand the underlying problem perfectly well but perform poorly because of the way the problem was presented.

The assessment records a negative result.

But what exactly has been measured?

The intended capability?

Or the person's ability to operate successfully within a particular interface?

When those two become confused, an interface can create a false negative about human capability.

That is not simply an accessibility issue.

It is an issue of assessment validity.

The diagnosis problem

Traditional accommodation systems often respond to this problem by asking people to demonstrate that they belong to a recognized category.

Provide a diagnosis.

Supply documentation.

Prove that you qualify for an adjustment.

Only then might the system change.

There are good reasons for formal accommodations in some circumstances. But the model contains an assumption worth challenging:

Why must someone prove that something is wrong with them before we remove a barrier that did not need to exist?

Not everyone has access to diagnosis.

Diagnosis can be expensive, geographically inaccessible, culturally dependent, or unavailable until adulthood. Many people spend years learning how to compensate for environments designed around neurotypical expectations.

Others never acquire a label at all.

Their difficulty with the system is nevertheless real.

A universal approach asks a different question.

Instead of deciding who is sufficiently different to deserve an exception, why not allow people to adjust aspects of an interaction that are irrelevant to what is actually being measured?

Need larger text?

Use it.

Need stronger contrast?

Change it.

Need less information on the screen simultaneously?

Reduce it.

Need information explained differently?

Ask.

Need the task presented one step at a time?

Do that.

None of those things necessarily changes the standard.

They change the route to the standard.

That distinction has become fundamental to our thinking:

Adapt the interface, not the capability standard.

AI changes what becomes possible

Historically, designing truly adaptive interfaces has been difficult.

Designers have had to anticipate user needs ahead of time and build settings, modes and accessibility features around categories of expected behavior.

AI creates another possibility.

A person can increasingly express an intention in ordinary language:

“Make this easier for me to scan.”

“Give me one thing at a time.”

“Explain that differently.”

“There's too much on this screen.”

“Show me the evidence before the explanation.”

“Don't simplify the idea. Simplify how you present it.”

The system can respond to the individual rather than forcing the individual to locate the correct setting or learn the architecture of the software first.

That is what we mean by intent-based UX/UI.

The person's intent becomes a primary interface.

The visual interface still matters enormously. Good information architecture, typography, interaction design and accessibility standards do not disappear.

But the interface no longer has to behave as though there is one correct way for a human being to interact with information.

AI as cognitive scaffolding

Our thinking was also influenced by something more personal: noticing how differently people experience AI itself.

There is increasing concern that heavy use of generative AI might cause people to disengage from thinking—to accept finished-looking outputs without making the decisions themselves.

That risk deserves serious attention.

But it is not the only way humans use AI.

For some neurodivergent people, AI can do almost the opposite.

Reading dense information, maintaining multiple ideas in working memory, organizing thoughts sequentially, returning to an interrupted idea, or locating the important point inside a large amount of text can consume enormous cognitive effort.

An AI conversation can reduce some of that peripheral burden.

The person can remain focused on the question.

Challenge an assumption.

Change direction.

Ask for another explanation.

Return to an earlier idea.

Keep thinking without having to hold the entire structure internally at once.

The AI is not necessarily doing the thinking instead of the human.

It may be removing obstacles that previously interfered with the human's ability to think.

That led us to another principle:

AI should reduce the cognitive burden of interacting with information without reducing human cognitive ownership of the decision.

The distinction is critical.

“Think for me” and “help me think” are not the same relationship with technology.

Removing the wrong friction

UX design has traditionally pursued frictionless experiences.

But we are learning that we need to be more precise about the kind of friction we remove.

Unnecessary interaction friction should disappear: confusing navigation, inaccessible typography, arbitrary conventions, excessive information density, poorly structured instructions.

But some cognitive friction is valuable.

A responsible AI might sometimes ask:

“Why do you believe that?”

“What evidence supports this?”

“There appears to be contradictory information.”

“Would you like to consider another interpretation?”

“Are you comfortable making this decision?”

That is not bad UX.

It may be how an AI system helps preserve human judgment.

Perhaps the objective is not frictionless AI at all.

Perhaps it is:

Remove interaction friction. Preserve judgment friction.

This problem extends far beyond education

Once we began looking for this pattern, we could see it almost everywhere.

Recruitment evaluates people through CV conventions, interviews, credentials and application systems.

Education evaluates people through standardized forms of teaching and assessment.

Government services require people to navigate bureaucracy and forms.

Workplaces privilege particular communication styles, schedules and behavioral norms.

Digital products assume particular ways of reading, concentrating and navigating.

People who fit those assumptions often never notice them.

People who do not may spend extraordinary amounts of energy adapting themselves, masking difficulties, developing workarounds or asking for exceptions.

And sometimes they simply disappear from the process.

The system then interprets their absence or poorer performance as evidence about their capability.

It may be nothing of the kind.

What this means for Adonea

Adonea began with a problem in employment: conventional recruitment systems frequently fail to represent what people can actually do.

But as we open the layers underneath that problem, something larger keeps appearing.

Human systems are full of proxies.

Credentials become proxies for capability.

Interview performance becomes a proxy for ability to perform the work.

Interface performance becomes a proxy for reasoning.

Conformity becomes a proxy for competence.

And people who do not fit the expected pattern can become invisible.

Adonea is being designed around a different assumption:

Human difference should not be treated as deviation from a default user.

That has implications for everything from opportunity discovery and evidence to assessment and interface design.

An Adonea interaction should increasingly be able to adapt around what someone is trying to accomplish without silently lowering expectations, changing the meaning of evidence, or making assumptions about the person's capability.

The same principle should apply to employers.

Different people need different amounts of context, explanation, structure and information density.

Universal design does not mean creating one interface that magically works identically for everyone.

It may mean creating a system capable of becoming appropriately different for each person while preserving the same underlying truth.

We are still learning

This is not an easy problem.

There are serious questions ahead about consistency, fairness, assessment validity, transparency, manipulation, AI error and how far adaptation should be allowed to go.

We do not yet have all the answers.

But one thing has become much clearer.

For decades, much of technology has effectively said:

Here is the system. Learn how to use it.

Human-centered AI gives us an opportunity to reverse that relationship:

Tell us what you are trying to accomplish. Let the system adapt around you.

And perhaps that is one of the more valuable things AI can do.

Not replace human difference.

Not normalize it.

Not diagnose it.

Not rank it.

But help remove the unnecessary barriers that have prevented us from seeing human capability clearly in the first place.

That is the direction in which our idea of intent-based UX/UI is beginning to emerge.