Aria Wang didn’t choose the food industry. It chose her twice — first when she graduated and took a job at ADM, one of the four largest agricultural trading companies in the world, and again after INSEAD, when the offer she took was in the same sector.
Six years at ADM, then business school, then a stint helping a biotech startup in Paris go from zero to one. Somewhere in there she stopped treating food as an accident of career path and started treating it as a choice.
“Food is the most universal form of communication,” she says. “Whatever you make, everyone gets to experience it. How much is a cup of bubble tea? Anyone can drink it. When you build a product you’re communicating with the world — the transaction is just the medium.”
That framing matters, because what she’s building now sits precisely at a gap between two industries that can’t hear each other.
The Thing That Surprised Me
After a month of intensive customer development in Chicago — the center of gravity for American food and beverage — Aria arrived at a conclusion that stopped me.
Not a single major food brand has a real AI strategy.
“The state of the industry right now is: they know the world is about to be changed by AI, but they don’t know what about them is going to change.”
These are companies with enormous budgets, world-class distribution, and products in nearly every household on the planet. They’re either hiring someone to write an AI strategy, or running internal initiatives that quietly go nowhere.
Two Industries That Can’t See Into Each Other
The reason isn’t money or will. It’s that food and tech are operating in separate epistemic worlds.
“Take a brilliant engineer out of Google and tell them to build a solution for a food company. Their first question is going to be — a solution for what?”
They have no model of how a product travels from consumer insight to formulation to sensory testing to regulatory review to shelf. They don’t know which steps are slow, which are expensive, or which have gone unquestioned for thirty years.
Flip it, and the problem is symmetrical. A food company’s organizational DNA has nothing to do with software. Its teams can’t imagine what AI would do for them because they’ve never seen the possibility space. Aria calls it a massive data silo between two industries — and notes that neither side is lazy or stupid. They simply cannot see into each other.
Her current strategy follows from this diagnosis. Before selling anything at scale, she’s working with academic institutions and industry executives on white papers that define a framework — showing food companies what their own workflows could look like. Occupy the conceptual space first. Commercialize second.
Service as Software, Not Software as a Service
The product decision follows the same logic.
Traditional SaaS sells a framework and depends on the user to input information, which the software then processes. Aria found that food industry executives responded to this exactly the way traditional business owners in every other sector do: If I buy this, do I have to use it myself? I don’t have time.
So she inverted it. Rather than software as a service, she describes what SKS does as service as software — turning professional work directly into something software delivers.
“Imagine I’m a company with money but no people. I can choose what resources I need and have agents deliver the work.”
The agents are organized by the org chart of a food company rather than by technical capability: marketing, R&D, consumer insights, formulation, sensory analysis, compliance. Fifty-six of them at last count, usable individually or composed into workflows — something closer to an agent store than a product.
One caveat she’s firm about: none of these run fully autonomously. Every agent is built on the assumption that a human provides the spark. AI executes and amplifies.
The 90 Percent Number
The capability customers want most is the digital consumer panel.
Traditional practice is to hire a third-party research firm and assemble real people for focus groups. Aria trains AI personas instead — each modeling the decision system of a consumer type — so brands can test product concepts against a digital panel before anything is manufactured.
Correlation with real-world results: 85 to 90 percent.
“AI is good at imitating human decision behavior. Give it enough context and the fit with real people is very high.”
She’s currently running a pilot with one of the largest snack brands in the world. It isn’t a wholesale replacement — 30 to 50 percent of the human panel runs in parallel with the digital one for validation. If it holds, the cost savings are better than tenfold.
The number worth dwelling on isn’t the 90 percent. It’s what produces the 90 percent. Anyone can call an API. Knowing which consumer dimensions actually drive purchase decisions, how a panel is structured, and what a brand manager will and won’t trust — that took ten years and can’t be prompted into existence.
Why Big Brands Make Boring Things
Ten years inside multinationals left Aria with a specific frustration.
“They have the best communication channels, the most resources, the most money — and what they deliver is basically noise.”
Her diagnosis is structural rather than creative. Large brands need scale to hit their economics, so they must serve what they believe is the broadest possible audience. Personalization gets stripped out — not because nobody wants it, but because operations and manufacturing can’t support differentiation.
“With AI, flexible production and personalization might actually become possible. When one person can become a team, the space for imagination is enormous.”
First Hackathon, First Place
While in Silicon Valley, Aria entered a Google DeepMind hackathon almost by chance. Three thousand participants. Her team took first in the semi-finals and second overall.
“When they called my name I was stunned. It was the first hackathon of my life.”
It also broke her image of the place.
“I used to think Silicon Valley was untouchable — so many brilliant people. Then I got there and found it genuinely open. They back people who haven’t finished university, who dropped out of high school. They actually believe things can happen. There aren’t that many constraints.”
She isn’t technical. She doesn’t claim to understand models at depth. What she has is a clear read on her own position:
“These people created these tools and gave me this capability. It used to be that you had to know code, know the language, to build anything. Now the thinking is — why shouldn’t anyone be able to turn an idea into a product just by speaking?”
Her entry point was mundane: using ChatGPT for MBA coursework in 2023, then talking to it constantly to find the edges of what it knew. “If you put me three years earlier, I couldn’t have done this. There’d have been no way to start.”
Stop Handing Yourself Biases
Before going to the US, Aria heard two things. One was encouragement. The other was doubt: You’re Chinese, you don’t have strong local resources, how are you going to do anything in their market?
She went anyway — first place at the hackathon, a solid customer pipeline out of Chicago, a co-founder and advisors found.
“Don’t hand yourself extra biases. Society already gives you plenty. Adding one you’ve endorsed yourself is pointless. There are opportunities in every environment.”
The Pattern Generalizes
I don’t think this is a food industry story.
Every traditional sector is sitting in the same position right now. They all know AI is changing the world. They don’t know how it changes them. The distance between those two sentences is where the next generation of companies gets built — not by the people with the best models, and not by the people with the deepest industry tenure, but by the small number who genuinely have both.
Most industries don’t have that person yet.
This article is adapted from 离线时间 EP15.