When I first started working on AI-powered products, I approached them the same way I approached any other digital product.
Define the screens.
Map the flows.
Design the happy path.
Handle edge cases later.
On paper, this made sense. It’s how we’ve designed software for years.
But very quickly, things started to feel… off.
The product worked. The AI responses were accurate. The UI was clean.
And yet, users still struggled to get value out of it.
That’s when it clicked for me:
I wasn’t designing a product for AI. I was designing a product around AI.
There’s a subtle but important difference between the two.
The “Around AI” Trap
Designing around AI usually looks like this:
- Take an existing product pattern
- Add an AI feature somewhere inside it
- Wrap it with UI, prompts, and guardrails
- Call it “AI-powered”
We treat AI like a component. A helper. A feature.
So we ask questions like:
- Where should the chat live?
- How many steps should the flow have?
- What happens if the response is wrong?
These are valid UX questions.
But they miss the bigger shift AI introduces.
AI doesn’t just change how users interact with a product.
It changes what the product fundamentally is.
AI Changes the Unit of Design
Traditional software is designed around actions.
Click this. Fill that. Move to the next screen.
AI-native products are designed around intent.
Users aren’t thinking:
“Which button do I press?”
They’re thinking:
“I want an outcome.”
That outcome might be:
- An Artifact
- A Plan
- A Decision
- A Summary
- A Recommendation
- A Next step
When we design around UI flows instead of intent, we force users to translate their thinking into our system’s structure.
When we design for AI, the system adapts to the user’s thinking instead.
This is where many AI products quietly fail.
The Product Is the Decision-Making, Not the Interface
Here’s a mental shift that helped me:
In AI products, the core product is not the interface — it’s the decision-making system.
The UI is just how users talk to it.
Once you see this, priorities change.
You start asking:
- How does the system understand intent?
- Where does human judgment still matter?
- How does the product recover from ambiguity?
- How do users steer outcomes, not just inputs?
This is very different from asking:
- Should this be a modal or a page?
- Should the button be primary or secondary?
Those questions still matter — but they’re no longer the center of gravity.
Designing With AI Means Designing With Uncertainty
Another thing I had to accept:
AI products are inherently uncertain.
Responses vary. Context shifts. User intent evolves mid-interaction.
Trying to eliminate uncertainty through rigid UX patterns doesn’t work. It just hides the problem.
Designing for AI means:
- Making uncertainty visible, not scary
- Allowing users to correct, refine, and guide
- Building trust through transparency, not perfection
This is less about polish and more about behavior. And it’s where designers can add real value.
What I’m Still Figuring Out
I don’t think there’s a single right way to design AI-native products yet. What I do know is this:
- Copying old UX patterns doesn’t work
- Treating AI as a feature undersells its impact
- Designing for intent requires unlearning as much as learning
I’m still figuring this out as I go.
But every time I stop designing around AI and start designing for it, the product feels simpler — and more powerful.
And that’s usually a good sign.
Final Thought
If you’re designing AI products today, don’t start with screens. Start with this question:
What decision is the system really helping the user make?
Design everything else around that.