I was talking to a colleague yesterday about a project, and somewhere in the conversation we got onto AX design.
I've been thinking about it since.
Most people in technology are familiar with UX, or User Experience design.
For years, UX has been about a fairly straightforward challenge: how do we make software intuitive, efficient and easy for people to use?
Designers have spent decades thinking about things like navigation, information architecture, interfaces, workflows, accessibility and how people move through a product.
At its heart, UX is about designing how a person interacts with software.
But AI is changing that relationship.
Traditional software generally waits for you to tell it what to do.
AI systems can increasingly interpret what you're asking, retrieve information, make decisions, use tools, ask questions and even take actions on your behalf.
The system isn't just responding anymore.
It's becoming part of the workflow.
And that creates a different design problem.
The question is changing
The question is no longer just:
How should a user interact with the software?
It becomes:
How should a human and an AI work together?
That's where we're starting to hear the term AX — AI Experience or Agent Experience.
There isn't yet one universally agreed definition of AX, and the terminology is still evolving.
But I think the underlying idea is important regardless of what we eventually decide to call it.
Once AI becomes an active participant in a workflow, designers have to think about questions that traditional UX didn't always have to deal with.
What should the AI do automatically?
When should it ask the user?
How much autonomy should it have?
How does the user know what the AI is doing?
What information is the AI using?
What evidence supports its answer?
What happens when it isn't sure?
How does a person correct it?
And, perhaps most importantly, when should the human take control?
These aren't just technical questions.
They're experience-design questions.
The interface hasn't disappeared
One thing I think is sometimes misunderstood about AX is that it doesn't replace UX.
Good AI products still need good interfaces.
People still need to navigate, review information, make decisions and understand what's happening.
But the role of the interface starts to change.
With traditional software, the interface is largely helping you operate the system.
With AI, the interface may also need to help you understand, delegate, supervise, correct and approve what the system is doing.
That's a significant shift.
Imagine an AI system that prepares a customer proposal.
The AI might search previous proposals, review customer information, identify relevant case studies, draft the response and recommend a particular approach.
The interface shouldn't simply present the finished proposal and say:
"Here you go."
A useful experience might also show:
- What the AI found
- Which information it used
- Where the information came from
- What assumptions it made
- Where there is uncertainty
- Why it is recommending a particular approach
- What the user needs to review
- What the AI is actually allowed to do
And then give the user meaningful ways to change, reject or approve the result.
The interface hasn't become less important.
Its job has become more complicated.
From operating software to supervising AI
This is one of the interesting differences between designing traditional software and designing AI systems.
With traditional software, we generally know what the system is going to do when someone takes an action.
Click this button.
The system performs this function.
Enter this information.
The system follows this workflow.
AI introduces much more variability.
There can be a gap between what the user asks, what the system understands, the information it retrieves, the reasoning it applies and the action it eventually takes.
That means the experience has to account for uncertainty.
It also has to account for trust.
If an AI system makes a recommendation, the user needs enough visibility to decide whether that recommendation should be trusted.
If the AI takes an action, the user may need to know what happened and why.
If the AI gets something wrong, the user needs a way to intervene.
And if the AI doesn't know, the experience should make it possible for the system to say so rather than simply producing a confident answer.
These are increasingly important parts of designing AI systems people can actually rely on.
AX doesn't mean giving AI more control
There's another distinction I think is important.
Designing a good AI experience isn't necessarily about giving the AI as much autonomy as possible.
Sometimes the right experience is for the AI to do everything automatically.
Sometimes it should make a recommendation and wait for approval.
Sometimes it should ask a question before continuing.
And sometimes it shouldn't be allowed to act at all.
The design challenge is understanding where those boundaries should sit.
That depends on the workflow, the consequences of getting something wrong, the quality of the available information and the level of human oversight required.
In other words, autonomy itself becomes something that needs to be designed.
What happens to UX?
I don't think UX is going away.
Far from it.
If anything, designing AI systems makes many of the principles of good UX even more important.
People still need clarity.
They still need consistency.
They still need to understand what is happening.
They still need to recover when something goes wrong.
But the designer is now dealing with another participant in the experience.
The software isn't simply a tool anymore.
In many cases, it is becoming a collaborator, assistant, decision-support system or agent.
That changes the relationship between the person and the product.
And it means we need to think about the experience of that relationship, not just the interface around it.
We're still figuring out what AX really means
I don't think AX is a fully formed discipline yet.
The terminology is still developing.
Some people use AX to mean AI Experience.
Others use it to mean Agent Experience.
And different teams are using the term in slightly different ways.
That's probably normal.
Technology often develops faster than the language we use to describe it.
But the design problem is already here.
As AI moves from something users interact with occasionally to something that actively participates in business workflows, we need to think much more carefully about how that interaction works.
Who decides?
Who acts?
Who approves?
Who can intervene?
What does the human see?
What does the AI know?
What happens when the two disagree?
Those are not questions we can leave entirely to the engineering team.
They're part of the product experience.
The bigger shift
For me, the interesting thing about AX isn't the terminology.
It's the shift in the underlying relationship.
Traditional UX asks:
How do we make software easier for people to use?
AI introduces another question:
How do we design a productive relationship between people and increasingly capable systems?
That relationship needs to balance capability with control.
Automation with oversight.
Speed with understanding.
And machine intelligence with human judgement.
I'm still working through what AX really means in practice, but I suspect it will become an increasingly important part of designing AI systems people can rely on.
Not because UX is becoming less important.
But because the thing we're designing is changing.
We're no longer designing only how people use software.
We're increasingly designing how people and AI work together.
Author
Vikram Katyani — Founder, IntelliMinds Digital.