Add a text box, connect it to a model, let people type what they want, and suddenly the product has an AI experience.
But I don't think every AI feature needs a conversation.
When I work on AI products, the first question isn’t “What should the chat look like?”
It’s:
What is the user trying to do?
Chat gives users freedom — and work
A conversational interface is useful when the user's request can take many forms.
They may need to explain context, ask follow-up questions, explore information, or change direction as they go. In those situations, forcing the interaction into a rigid form can make the product harder to use.
Chat gives people room.
But that freedom also creates a new responsibility for the user.
They have to understand what the AI can do, decide what to ask, phrase the request, evaluate the response, and figure out what to do next.
Sometimes that's exactly the right interaction.
Sometimes we're making the user design their own workflow.
If the possible actions are already known, show them
Imagine an AI feature that can do four things to selected text:
- translate it
- rewrite it
- make it more formal
- summarize it
You could give the user a chat box and expect them to type:
“Can you make this more formal?”
But the product already knows the available actions.
In that situation, showing Translate, Rewrite, Formal, and Summarize directly can be faster and clearer than asking the user to prompt the system.
The AI is still doing the work.
It just doesn't need to look like chat.
This is an important distinction in AI Product Design: the model and the interface are two different decisions.
Using an LLM doesn't automatically make conversation the right interaction model.
Structured UI can reduce uncertainty
Traditional interfaces are good at showing possibilities.
Buttons show actions.
Forms show required information.
Filters show available choices.
Progress indicators show where you are.
AI chat often hides those boundaries behind an empty input.
That can be powerful when the space of possible requests is genuinely open. But when the task is constrained, hiding the structure can create unnecessary uncertainty.
The user starts wondering:
- What can I ask?
- How should I phrase it?
- Did the AI understand what I meant?
- What happens if I ask differently?
Sometimes the best AI UX is simply making more of the system visible.
Chat works better for exploration
There are situations where I would strongly consider conversation.
For example, when someone needs to:
- investigate information across several sources
- refine a question through follow-ups
- explain a situation with a lot of context
- explore possibilities before deciding what they need
- move between related questions without restarting a workflow
Here the conversation itself carries useful context.
The interaction isn't just command → result.
It's a process of understanding.
That's where chat becomes much more interesting as a product interface.
The answer is often a hybrid
I don't think the choice is usually “chat or no chat.”
Some of the more useful AI experiences combine conversational and structured interaction.
A user might begin with a question, receive an AI-generated answer, and then use explicit controls to verify information, change a parameter, select a source, approve an action, or continue a workflow.
Or the opposite can happen.
The product starts with structured choices and opens conversation only when the user's situation becomes too complex for those choices.
That gives the AI flexibility without forcing every interaction through a prompt.
Designing the response matters too
Choosing chat is only the beginning.
An AI response isn't the same as a normal message.
It can be incomplete, generated from specific sources, uncertain, incorrect, interrupted, regenerated, or affected by context the user cannot see.
The interface therefore needs to answer questions that ordinary messaging interfaces don't:
- Where did this information come from?
- Can I verify it?
- Can I stop or retry this?
- What context is the AI using?
- What happens when something goes wrong?
These aren't visual details added after the model works.
They're part of the product experience.
Start with the task, not the chat box
When I'm thinking through an AI feature, I find it more useful to start with the user's task and work outward.
What are they trying to accomplish?
How much freedom does that task actually require?
Which decisions can the interface make visible?
Where is AI useful?
Where does the user need control?
And only then:
Should this part be conversational?
Sometimes the answer is chat.
Sometimes it's a form, an inline action, a review panel, a set of controls, or a combination of them.
AI Product Design isn't about putting a conversation on top of a model.
It's about finding the interface that makes the model useful.