Open any new AI product launched in the last two years and there's a good chance you'll be staring at a text box. Type your prompt, get your answer. The chat interface has become the default UI pattern for AI tools the same way the hamburger menu became the default for mobile apps: not because it's the best solution, but because everyone else is doing it.
UX designer Katya Korovkina calls this "chatbot-first thinking" in a recent piece for UX Collective, and her core argument is hard to argue with. Most tasks people do with AI tools don't naturally fit a conversation. Editing a document, sorting data, adjusting an image, managing a project - these are spatial, visual tasks being crammed into a linear text thread.
The Chat Box Isn't Always the Answer
Think about how you actually use AI in your work. When you ask ChatGPT to rewrite an email, a chat interface makes sense. But when you're using an AI-powered design tool, spreadsheet assistant, or project manager, the chat box becomes an awkward middleman between you and the thing you're trying to do.
The better approach, Korovkina argues, is treating AI as an invisible layer that powers the application rather than a feature bolted onto it. Notion's AI is a decent example of this: it lives inside the editor where you're already working, not in a separate chat panel you have to context-switch into. Canva's Magic tools work the same way - the AI operates within the design canvas, not alongside it.
This matches what I've seen testing dozens of AI tools over the past year. The ones that feel most natural aren't the ones with the fanciest chat interfaces. They're the ones where you barely notice the AI is there.
Generative UI Is the More Interesting Bet
The piece highlights a trend called "generative UI" where instead of designers drawing every possible screen, they design components and systems, and the AI assembles them based on what a specific user needs at that moment. Your dashboard looks different from mine because the interface itself adapts.
This is already showing up in tools like Framer and some newer prototyping platforms. It's still early, but the idea makes practical sense: instead of building one interface and hoping it works for everyone, let the AI customize the layout, information density, and feature prominence per user.
The catch is trust. Users need to understand why the interface looks the way it does, and they need confidence the AI isn't hiding important options or making bad assumptions about their intent.
Trust Markers Over Magic
Korovkina points to "trust markers" as a design pattern that more AI tools need to adopt. These are visual cues that explain what the AI did and why - showing its sources, flagging its confidence level, or letting users see the reasoning chain behind a suggestion.
This is the part most AI tools still get wrong. They present outputs as finished answers with no indication of how reliable they are. A tool that says "I'm 60% confident about this answer, and here's the data I used" is more useful than one that presents everything with equal authority, even if the confident-sounding version feels more polished.
The broader point is worth sitting with: we're still in the phase where AI tools copy each other's interfaces rather than designing for their actual use cases. The chat box was a reasonable starting point. Two years in, it shouldn't still be the default.