The hard part of AI isn’t making things. It’s choosing.

Opinions expressed by Digital Journal contributors are their own.

Most new AI products promise to do the work for you. Jingyi Zhu, a product and UX designer, is more interested in what happens next: the point where a person has to look at what the software produced and decide whether any of it is worth keeping. 

That question runs through her work. As AI tools get better at generating almost anything, she argues, the scarce human skill becomes judgment, the ability to tell what is good, what fits, and what to cut. Her aim, she says, is to make that judgment easier, not to remove the person from the process. In 2026, two products she worked on won Red Dot Awards. 

She describes treating a screen full of AI options the way an editor treats a rough draft: material to be cut down and made clear. The test she applies to a new AI feature, she says, is whether it helps a person make a better decision or simply makes the decision for them. 

Castor is one example. The AI content studio, a 2026 Red Dot Award winner, helps creators turn a personal brand into a steady stream of social media, and Zhu was the designer behind its experience. The part that best shows her thinking is the “Digital Twin”: an AI persona that learns a creator’s voice and drafts posts in it, while the creator keeps final approval over everything that goes out. Nothing publishes on its own. Each draft arrives as a suggestion the creator can keep, rewrite, or discard, with the reasoning behind it in view, and when the system is unsure, it says so rather than dressing up a guess as a finished post. The AI does the drafting; the person decides what actually sounds like them. 

“When a tool can make a hundred versions of anything, making more isn’t the hard part anymore,” she says. “The hard part is knowing which one is right. That’s what I try to make easy.” 

She designed PhotoG on the same principle. The agentic AI marketing platform, another 2026 Red Dot Award winner, could have been four separate AI tools; instead Zhu brought them into a single two-panel workspace, so one marketer runs the whole pipeline with a sign-off at every handoff. The layout keeps the AI’s work on one side and the person’s controls on the other, so approving, editing, or redirecting a step never means leaving the flow. In both products, the system does the heavy lifting, but a person stays in the loop wherever judgment matters. 

That instinct comes from years of building digital products, and from watching how people actually behave when a tool starts acting on its own. She is skeptical of interfaces that ask users to trust a system without question. “Blind trust isn’t the goal,” she says. “People should know when they can rely on the system, and when they need to look more closely.” 

The issue is practical as well as conceptual, she notes. In marketing and creative work, an AI that makes an off-brand choice can cause damage before anyone notices, and a single tone-deaf post published automatically can set back months of work. Her designs are built around that moment, when a person needs to catch that something is off and pull it back. Often, she says, that means having the tool pause for a review before something goes out, not after. 

Zhu expects the question to grow more common as AI tools take on more of the production and leave people with the decisions. In her view, the interface becomes the place where a person’s judgment meets what the machine can do, a problem she says the field is only starting to address. The tools she wants to build, she says, are the ones that keep that judgment central rather than optional.