Zui Chen on Designing Interfaces When AI Takes the Wheel

dscene magazine has published a feature interview with product designer Zui Chen titled "How Product Designer Zui Chen Designs Interfaces When AI Acts." The piece takes apart the design choices behind software that no longer waits for a click.

Zui Chen, a working product designer, sits at the center of the article. The piece frames the design problem as one of agency. When a system acts on its own, what does the interface owe the user? That question has grown louder as agent-style features keep shipping across consumer software.

The original article lives in dscene magazine. Readers who want the full interview will need to follow the source link. The original piece sits behind a standard news gate.

What does "AI acts" actually mean here?

In practice, the term points to systems that perform tasks without a per-action prompt. A tool that drafts replies. An assistant that books meetings. An agent that runs a multi-step workflow. Each one shifts the user from operator to supervisor.

Interfaces built for that shift don't always look different at first glance. The work resembles cockpit design more than screen design — most of the surface stays quiet, and most of the action happens out of view. The cursor stops being the main character. Status indicators, confirmation moments, and override controls do more of the work.

What pressure points does the title flag?

Attribution comes first. When AI does something, the interface must say so without burying the action in noise. Control follows. Users need a way to step in, pause, or redirect. Feedback rounds it out. The system must show its hand before, during, and after each move.

Chen's title alone signals the work stays grounded. Product designer, not theorist. The article's framing suggests it walks through specific interface decisions. The kind that show up in design files and shipped releases, not conference slides.

What does it mean for product teams shipping AI features?

For teams shipping AI features in the next quarter, the practical questions stay small and stubborn. How do you signal that a model is about to act? Where does the stop button live? What does "undo" mean when the action already left your system and hit someone else's inbox? The interview, on the title and framing, treats those questions as the job itself — not as edge cases.

The publication lands in a year crowded with agentic-product launches. Designers across the industry now argue over how much autonomy an interface should grant by default. The interview offers a practitioner-level counterweight to the usual vendor write-up.

Why does this interview land now?

The piece matters because most public writing on agent design still comes from the model vendors themselves. A designer-led account of what those interfaces look like in the wild offers something different. It also gives product teams a vocabulary for the moments that matter most: the pause before an action, the reversal after one, the quiet acknowledgment that an autonomous step happened at all.

The full feature runs in dscene magazine. As agent-style products move from demo to default, the design vocabulary around them remains unfinished — one frame at a time.