I built my first interface on the assumption that everything the user needed had to be visible. Twenty years in automotive taught me to fear the hidden control — if it mattered, it went on the screen, one glance away. That instinct served me through enterprise dashboards, 3D tools, a hundred products. And then it started to fail.
The screen was never the point
A screen shows you everything at once and asks you to find what matters. For a long time that was the best we could do: lay out every option, label it well, and trust the user to navigate. But navigation is work — work we'd quietly offloaded onto the person using the thing.
When a system can understand intent, that calculus changes. The user says what they want; the interface figures out the path. The screen stops being a map you read and becomes a place you arrive.
A screen shows everything and asks you to find what matters. A conversation withholds — and that restraint is the design.
What a conversation demands that a screen doesn't
Designing a conversation is harder, not easier — precisely because there's less to see. A button either affords a click or it doesn't. But a system that answers has to decide what to say, how much, in what order, and when to stay silent. Those are editorial decisions as much as visual ones.
In practice, I found the work moved into three places that never used to be 'design' at all:
- The prompt — the tone, the guardrails, the personality the system speaks with.
- The fallback — what happens when the model is unsure, wrong, or asked something out of scope.
- The trace — how a user confirms that an answer is real and where it came from.
const assistant = { persona: "precise, calm, never oversells", onUnsure: "say so, offer the source", maxWords: 60 }
Designing the parts you can't see
This is the uncomfortable part for a visual designer: much of the quality now lives in behavior nobody can screenshot. Whether the assistant admits uncertainty. Whether it remembers what you said three turns ago. Whether it knows when not to help. You can't put those in a Figma frame, but users feel them within seconds.
So I started prototyping in language before pixels — writing the conversation as a script, reading it aloud, cutting every line that didn't earn its place. It turns out 'less is more' survives the jump from screens to dialogue completely intact. If anything, it matters more.
Where this leaves the craft
I haven't stopped designing screens — I'm doing it right now, on this site. But the screen is no longer where the thinking ends. It's the calm surface over a system that listens, reasons, and occasionally speaks. The best interface I can build today is one that shows less, because it finally understands enough to know what you actually need to see.