Most AI products I’m asked to look at have the same front door. A text box, a blinking cursor and a line that says “Ask me anything.”
It’s easy to see why. Chat is the fastest way to put a model in front of users. The API returns text, so you show text. Ship it and see what people type.
The trouble shows up a few months later. Usage is lower than the demo suggested. The people who do use it type the same few requests every day. And the founder starts to feel their product looks like every other AI dashboard out there.
Over the last year, five AI founders have booked a call with us asking for some version of the same thing: an AI product that isn’t a chat window. After 20 years designing products, and 50+ AI and crypto founders through the studio, it’s the brief I enjoy most. So here’s how we approach it.
Why do so many AI products default to a chat interface?
Because it’s the easiest thing to build. That’s a different question from what’s easiest to use.
A chat box hands the user the hardest part of the job. They have to work out what to ask, phrase it well and judge whether the answer is right. Good software has spent decades taking that work off people with menus, defaults, smart suggestions and a button for the obvious next step. A blank prompt puts it all back.
Chat also hides what your product can do. A new user staring at an empty box can’t see where the edges are. So they ask something small, get something average, and leave.
What should an AI interface look like instead?
Start from the job. Early on I ask founders one question: what is your user in the middle of doing when the AI helps them? The answer usually points to the pattern. These are the ones we reach for most.
AI inside the screen they already use. If someone’s writing an email, updating a deal or reviewing a trade, the AI shows up right there as a suggestion they can accept, tweak or ignore. No context switch and no prompt to write. A lot of the best AI features barely look like AI.
A review queue. When the AI works in the background, the interface becomes an inbox of finished work waiting for a yes. Drafts to approve, records it updated, leads it scored. The user’s job moves from doing to checking, and the design work is making checking fast and trustworthy.
A canvas. Some work has a shape. Retell, the YC-backed voice agent platform, is a good example. We rebuilt their design system and core UX flows, and the people building voice agents there work on a canvas where every step of a call is laid out at once.
Structured output. When the model returns something with structure, show the structure. A comparison is a table. A plan is a timeline. A set of options is a row of cards with the trade-offs on them. Paragraphs make people do the formatting in their head.
Ambient status. Agents that run for minutes or hours need a home on the screen that says what they’re doing, what they’ve finished and what they need from you. Think of a file upload’s progress bar, where the file is a piece of work.
When is chat still the right interface?
I’m not against chat. It earns its place when intent is genuinely open-ended, when someone is exploring, or when a power user knows exactly what to ask. It’s also a great escape hatch for the request your interface didn’t see coming.
The mistake is making it the front door. We usually keep a chat or command input somewhere in the product, then design the main experience around the three or four jobs that make up most of the usage.
How do you design an AI product that feels AI-native?
We look at real usage first. If the product is live, the chat logs are the best brief you’ll ever get. The requests people type every day are the buttons, defaults and suggestions your interface should already have.
We design for the wrong answer. Models get things wrong, so the interface has to make that cheap to catch and fix. That means showing sources, making edits easy and being clear about what the AI did and what the user did.
We prototype in code. A static mock-up can’t show you how an AI interface feels, because the feel lives in timing, streaming, loading states and what happens when the output is long or strange. We build the core flows as working prototypes early, with real model output, so the team reacts to the real thing.
And we treat response time as a design decision. If an answer comes back in a second, it can sit inline. If it takes a minute, it becomes a job the user can leave and come back to. Making that call early saves a lot of rework.
A quick test for your own product
Open your product, find the chat box and ask yourself:
What are the five things people type into it most?
Could any of those be a button, a default or a suggestion on the screen where they’re already working?
Does a new user know what the product can do before they type anything?
When the AI gets it wrong, how many clicks does it take to fix?
If the honest answers are uncomfortable, you’re in good company. Most of the AI products we see still have the chat box doing jobs the interface should be doing.
Let’s compare notes
I’m writing here because I want to hear how other founders building with AI are handling this.
So here’s my question for you: where in your product is chat doing a job a better interface should be doing? Leave a comment or reply to this email, and I’ll tell you what I’d try.
If you’d rather we looked at it together, this is the work we do at Halaska Studio, the product design partner for funded AI and crypto teams. See how we work.
Do AI products need a chat interface?
No. Chat suits open-ended tasks and power users, but most AI products work better when the AI sits inside the screens where the work already happens, with chat kept as a fallback.
What are the alternatives to a chatbot UI?
The patterns we use most are inline suggestions, review queues for work the AI did in the background, canvases for work with a structure, structured output like tables and cards, and status views for agents that run for a long time.
What is AI-native product design?
AI-native product design shapes the whole product around what the AI does well and where it fails, instead of adding a chat box to an existing app. It covers the interface, how the AI’s work is shown and checked, and how fast each answer needs to come back.
How do you prototype an AI interface?
We build the core flows as working prototypes in code with real model output. Static mock-ups miss the parts that decide how an AI product feels: streaming, waiting, long answers and wrong answers.
Who designs AI products that go beyond chat?
Halaska Studio is a product design studio for funded teams, directed by Chris Halaska from Dubai. We design AI and crypto products as an ongoing partner, with senior designers and a founder who has led product design at Google and Immutable.
the review queue idea made sense to me, moving the user from doing to just checking is a nice way to see it.