When I first encountered agentic AI systems, my instinct was to lean on familiar UX patterns.
Define the flow.
Constrain the steps.
Make the system predictable.
That’s how we’ve always designed reliable software.
But the more I worked with agent-based systems, the more I realized something uncomfortable:
Traditional UX patterns don’t just struggle here — they actively get in the way.
Not because the patterns are bad.
But because the assumptions behind them no longer hold.
Traditional UX Assumes a Stable System
Most UX frameworks are built on a simple premise:
- The system is deterministic
- The user performs actions
- The system responds predictably
In that world, good UX is about:
- Clarity
- Reducing cognitive load
- Guiding users through known paths
Agentic AI systems break this model entirely.
Here, the system:
- Reasons
- Plans
- Delegates
- Adapts
It doesn’t just respond — it acts.
And once the system starts acting on behalf of the user, the idea of a fixed flow starts to collapse.
Flows Don’t Work When the System Thinks
One of the first things I noticed was how quickly carefully designed flows became irrelevant.
In agentic systems:
- The same intent can lead to different paths
- Steps may be skipped, reordered, or expanded
- Outcomes emerge rather than execute linearly
Trying to force this into a step-by-step UX often results in:
- Unnecessary confirmations
- Rigid checkpoints
- UI that exists purely to “feel safe”
Ironically, this often makes the experience less trustworthy.
The user isn’t confused because the system is complex.
They’re confused because the UI is pretending the system is simpler than it actually is.
Control Shifts From Screens to Boundaries
Traditional UX gives users control through interface elements:
- Buttons
- Forms
- Toggles
- Screens
Agentic AI shifts control to something more abstract:
- Constraints
- Intent
- Guardrails
- Correction
Users don’t want to control how the system works. They want to control what it’s allowed to do.
This is where old UX habits break. We keep designing more UI to regain control, when what’s actually needed is:
Better affordances for steering
Clearer ways to intervene
Simple mechanisms to say “this is not what I meant”
Control becomes conversational, not procedural.
Predictability Is Replaced by Recoverability
Another assumption traditional UX relies on is predictability.
If the system behaves predictably, users feel safe.
Agentic systems don’t offer that luxury.
Responses vary. Plans evolve. Context changes mid-interaction.
Trying to make agentic AI predictable is a losing battle.
What does work is designing for recoverability:
Can the user correct the system easily?
Can they backtrack without penalty?
Can they refine intent without restarting?
This is less about perfect UX flows and more about humane system behavior.
The Designer’s Role Quietly Changes
This shift forced me to rethink what “good UX” even means here.
It’s less about — polishing screens, optimizing flows, reducing steps
And more about:
Shaping system behavior
Deciding when humans step in
Designing trust through transparency
In agentic systems, designers are no longer just interface designers.
They’re shaping how intelligence behaves around humans.
That’s a very different responsibility.
What I’m Still Unlearning
I still catch myself reaching for familiar patterns:
“Should this be a wizard?”
“Do we need a confirmation modal?”
“What’s the happy path?”
Sometimes those questions help. Often, they don’t.
What helps more is asking:
Where might this system misunderstand intent?
How does the user regain control when it does?
What does trust look like after something goes wrong?
I’m still learning to design for that reality.
Final Thought
Agentic AI doesn’t break UX because it’s too complex.
It breaks UX because it exposes assumptions we didn’t realize we were making.
When systems start acting, planning, and deciding,
design has to move upstream — from screens to behavior.
That’s uncomfortable.
But it’s also where the most interesting work now lives.