She uses AI every week to explore ten creative directions in the time one used to take. She also packs boxes by hand on a London production floor. Experiential designer Poppy Haryadi argues that the biggest risk of AI in design isn't replacement; it's a beautiful lie.

There is a version of the AI-and-design conversation that has become almost a genre. In one corner, the evangelist, for whom every workflow is about to be transformed and every skeptic is a Luddite. In the other, the refusenik, for whom the tools are theft, slop, or both, and the only honorable position is abstention. The debate is loud, moralized, and, for anyone who actually makes physical things for a living, strangely beside the point.

Poppy Haryadi belongs to neither corner, and her working week explains why. Haryadi is a brand and experiential designer based in London, whose portfolio runs from brand activation builds for American Express at Rockefeller Center and Art Basel Miami to luxury beauty mailers for NARS, Carolina Herrera and Neutrogena, and gifting for Netflix, Spotify, Tinder and Ministry of Sound. On any given morning she may be using AI tools to explore concept directions at a speed that would have been unthinkable three years ago. And on any given afternoon she may be standing next to a printer at Bespoke-ly / Not Another Bill, the London studio where production happens in-house, watching how a foil actually sits on a particular stock, or, occasionally, packing the boxes herself.

"Things are changing fast, and AI tools have become a natural part of how I work," she says, without ceremony. "I use them in the early concepting phase, not as the work itself, but as a way to explore ideas quickly and pressure-test a direction before I commit."

It is precisely because she lives at both ends of the process, the accelerating digital front end and the stubbornly physical back end, that her account of what AI is doing to design is more useful than most. And her central warning is not the one the industry keeps arguing about.

What actually changed

Start with what the tools have genuinely transformed: the front end of creative work has been rebuilt. Haryadi is unsentimental about it.

"Concepting will get faster and wider," she says. "AI already lets me explore ten directions in the time it used to take to explore one, and pressure-test them before committing to anything. In five or ten years I'd expect a lot of that early visualization to be almost instant."

For a designer whose process leads with moodboards and visual direction, getting a concept to where it "feels right for the brand before it gets over-explained," as she puts it, this is not a threat but an amplifier. The exploratory phase that once consumed days of pulling references and building mockups now compresses into hours. Directions that would have died as hunches, too expensive to visualize, can be tested and discarded cheaply.

And she is clear-eyed about what that abundance does to the economics of her profession. "That's freeing, because it means less time proving an idea can exist and more time spent on whether it should," she says. Then comes the observation that separates her from the evangelists: "I also think judgment becomes more valuable, not less, because when everyone can generate options, knowing which one is actually right for the brand is the scarce thing."

This is a quietly radical reframing. The industry's anxiety has fixated on generation, who makes the image, the render, the option. Haryadi's point is that generation was never the valuable part. In a world where every junior designer, every client, every brand manager can produce forty plausible directions before lunch, the entire weight of the profession shifts to selection: the trained instinct that recognizes which of the forty actually serves the brand and the person on the receiving end. Her own version of that instinct, she has said elsewhere, is almost physical, "Taste is mostly a kind of discomfort. You feel the off note before you can explain it." AI can generate the notes. It cannot yet feel the off one.

The beautiful lie

But it is Haryadi's second observation that deserves to reshape how the industry talks about these tools, because it names a risk almost nobody else is naming.

"There's a real risk if we lean on it carelessly, and it's one I think about a lot with mailers and experiential specifically," she says. "AI can make something look extraordinarily polished, flawless, in a way the physical version sometimes just can't match, because production is still human and the end product is a real object someone holds in their hands."

Here is the trap, laid out plainly. Generative tools do not merely visualize ideas faster; they visualize them better than reality. The AI render of a packaging concept has no print tolerances. Its foil never bubbles, its ink never sits heavy on the wrong stock, its structural folds never fight the grain of the board. It is a physics-free preview of an object that will, eventually, have to obey physics.

"If you let the AI render run too far ahead, you create an expectation the actual build can't meet," Haryadi says. "The screen promises something the material can't deliver, and that gap is a genuine problem."

Anyone who has worked in production will recognize the scenario she is describing, because a milder version of it has always existed: the client who approved the retouched deck visual and is disappointed by the printed proof. What AI does is industrialize that gap. When the approval image is generated rather than mocked up, it is untethered from manufacturability by design. The client falls in love with something that cannot exist. The designer inherits the impossible task of closing the distance. And the finished object, the real thing, in a real hand, reads as a failure not because it is badly made, but because it was compared to a lie.

Her conclusion is neither ban nor surrender, but discipline: "For the kind of work I do, I think it always has to be a combination of both, the tools to explore and imagine, but grounded constantly in what can truly be made, printed, packed and touched. The render has to stay honest to the reality of production."

The antidote is on the production floor

If the disease is renders that outrun reality, Haryadi's antidote is biographical: get closer to reality than the render can.

Her current chapter at Bespoke-ly / Not Another Bill has given her something most designers never get: production in the same building. After years in New York working with outsourced fabrication, she now designs within arm's reach of the people and machines that make the work, and the education has been physical. "When you're close to how things are actually made, you see everything the screen hides," she says. "How an ink or a foil sits on one surface is completely different from how it sits on another... Paper stock and weight change how a print reads entirely. On screen you can convince yourself of almost anything, but in your hands, the truth of it is obvious."

She has even embraced the least glamorous station in the workflow: the packing table. "At first that's not what you picture a designer doing," she concedes. "But watching how a piece actually gets assembled, how it's placed and folded and set into the box, teaches you things the artwork alone never could... If it isn't put together well, the whole thing loses its magic, no matter how good it looks on paper."

The payoff is a kind of embodied quality control that runs in the opposite direction to the AI trap. "After doing it for a while, that knowledge moves into your hands," she says. "Now I mostly design with it built in from the start; I can feel whether something will work in production before it gets there, almost without thinking about it."

Read those two halves of her practice together, and a model emerges for what the AI-era designer actually looks like. Not the prompt artist, generating unmoored beauty. Not the abstainer, refusing the speed. But something closer to a translator with one hand on each side: fluent enough in the tools to exploit their exploratory power, and literate enough in material reality to keep every render honest. The machines widen the funnel of what can be imagined; the production floor narrows it to what can be held.

What stays irreplaceable

Which brings the argument to its foundation, the part of Haryadi's work she believes no tool will touch, no matter how the next decade unfolds.

"You cannot AI your way into how it actually feels to walk into a space, or to open something with your hands," she says. "The scale of a room, the light, the texture of a material, the way a crowd gathers, the small imperfections that make something feel real, none of that survives on a screen. Our work exists to make people feel something in their body, in a real place, with other real people. The tools can help you plan that, but they can never replace it, because it only exists once it's actually built and someone is standing inside it."

She knows this from the inside. Her formative professional moment came standing in a build she had helped design, the American Express Gold Card x Luar launch at Rockefeller Center during New York Fashion Week, watching a plaza full of people have a genuinely good night. What struck her was that the design itself disappeared into the experience it created. No render, however flawless, contains that.

And here Haryadi lands on the most counterintuitive note in her whole position: she believes the technology that generates infinite screen-perfection will make the physical more valuable, not less.

"So much of our lives is on a screen now that being somewhere real, holding something real, actually being in the room, starts to feel like the rare part, and that's the thing people remember," she says. "So I don't really worry about the tools replacing that side of the work. If anything, they take care of the exploring and the guessing, which frees me up to focus on the part that was always the point: what it actually feels like to be there."

It is a strangely hopeful place for an AI argument to end, not with a defense of human labor, but with a defense of human presence. The tools will keep getting better at showing us things. Haryadi is betting her career on the difference between being shown something and standing inside it. On current evidence, a top five dining destinations brand lounge here, a sunscreen box that only tells the truth in sunlight there, it looks like a good bet.

About Poppy Haryadi (Noer Giovanda Poppy Haryadi). She is a brand and experiential designer based in London, creating brand activations, packaging, and gifting for clients including American Express, NARS, L'Oréal, Carolina Herrera, Neutrogena, Netflix, Spotify, Tinder, and Ministry of Sound. Trained first in interior architecture in Indonesia, she built her practice in New York, contributing to landmark builds at Rockefeller Center and Art Basel Miami, before bringing that standard to London's gifting market.

Haryadi is a distinctive voice on AI in design: an early adopter who uses generative tools throughout her concepting phase to explore and pressure-test directions, while working hands-on with in-house production at Bespoke-ly / Not Another Bill. Her argument that flawless AI renders risk promising what physical production can't deliver, and that judgment, material literacy, and real-world presence become more valuable as generation gets cheaper, bridges the gap between AI evangelism and refusal.