For this episode I brought together two very different people to talk about marketing in the AI era.
Yichen founded CapgoAI, which does AI-driven SEO and GEO for AI and software companies. Harvard MBA, a classic AI-industry founder.
Ashley founded Sukoon, a body-care brand in Singapore: shower gel, refills, fragrance. She is a one-person company. She launched last December and sold about 3,000 orders in eight months. She doesn't work in AI, but the way she uses it is more interesting than many AI companies.
The theme was AI-native marketing: what actually changes and what doesn't, across three parts, workflow, taste and pattern.
A year of talking to seven bots
For Sukoon's first year, from brand positioning to supply chain to product development, it was Ashley and ChatGPT. She built a virtual team inside it, one for reviews, one for strategy, one for product.
Then, at launch, she said something I didn't expect:
"I suddenly lost the desire to express. I really wanted to hear what real people thought of the product."
So she made a decision that looks anti-AI: all in on offline. Seven fairs and pop-ups in eight months, selling one order at a time and watching every stranger face to face. She wasn't looking at sales data. She was reading micro-expressions: why someone stopped at the stall, their first reaction to the scent, which line made them ask the price, how they reacted to it.
"I might repeat that process hundreds or thousands of times."
Those expressions were her database, and she was the processing engine. AI couldn't give her that.
Perception can't be replaced
What has AI taken over? Finance and legal: download and upload the reports and AI handles about 80% of the analysis. Images: she shoots first and AI does a second pass.
What never will? Judgment.
"Whether it's a product decision or a market call, in the end it rests on my experience and perception. AI can't replace that, because perception can't be replaced."
Choosing the bottle, she walked into a factory with three or four hundred options and knew hers from twenty metres away. Choosing the scent, she asked herself: if Sukoon were a person, what moment would make it feel most alive? Her answer was peeling a mandarin in the February sun, the moment the mist bursts from the peel.
None of that comes from data. Eight months of selling in person showed that what she thought was good, the market thought was good too.
AI draws conclusions even when it doesn't know enough
Yichen spoke from the agency side. Much of CapgoAI's SEO work is automated now. The one thing that isn't is working with the client.
"The client may not be optimising for traffic. They may be optimising for the click-through rate on the five keywords their boss cares about."
AI can't get that context. Worse, it concludes anyway. It doesn't say it lacks the information to judge. Ask it which keywords a site targets and it may read only the URL, assume it has read the page, and answer wrongly. You have to force it to actually read.
Too much context fails too. People abstract what matters. AI gets confused. Two sentences from the boss outweigh a hundred from the team, and AI can't tell.
The biggest waste isn't slowness, it's building the wrong thing
To Yichen, taste is judgment: deciding what to do. He shared a Harvard professor's view on how to train judgment: make more decisions you're accountable for. The professor had laid off forty or fifty people across his companies and remembered every name and every reaction, while he couldn't recall any of the hundred-plus layoff cases he had read.
Early on, Yichen thought engineering speed was the company's biggest loss.
"Looking back, the biggest waste was all the features I built that turned out to be useless."
Judgment is worth more now because execution is cheap. Building something is much easier than before. Building the wrong thing costs just as much as it did.
Zero times ten is still zero
On go-to-market, Yichen thinks enterprise B2B hasn't fundamentally changed. In person matters more than ever, because everyone's inbox is full of automated outreach. Where things have changed is consumer and product-led growth, where AI amplifies strong individuals.
"If you're a one, AI multiplies you by five or ten. If you're a zero, you're still zero."
Before guns, fights were won by martial skill. With guns, it's who has the better gun and the better aim. With AI, it's how well you use it, plus your judgment and taste. You still have to be better than average to win customers.
Using AI to distil her own intuition
Ashley doesn't use AI to replace her judgment. She uses it to understand it. She asks ChatGPT why she chose green citrus peel, why a woody base. AI helps her pull frameworks out of those instincts: she was responding to people's need for certainty in the AI era, and certainty felt like something direct, immediate, needing no thought.
She also built a version of herself thirty years from now and works backwards from that end state to check each decision against it.
But tools change so fast they became a burden. Her rule now is to let things settle, go deep in one tool, and stop chasing new ones.
A tenth of the salary, a tenth of the time, twenty years of judgment
Her team structure is what I most want to remember: a one-person company, a matrix of AI agents, and four part-time specialists, three of them with more than twenty years in the industry. She pays them a tenth of a normal salary for a tenth of their time.
"It's like fine dining. You pay a lot for the set menu, but what you're really buying is the judgment the sommelier and the chef built over twenty years."
She spent the first seven months selling offline, one order at a time, and didn't rush to scale. Asked why she didn't go to livestreaming in month two, she said that if you sell 10,000 orders you get 1,000 reviews. What do the other 9,000 people think? Not knowing scared her more.
At a trade show in Bangkok, Sukoon was the youngest brand in the hall, seven months old, and the only one to sell out.
One good video beats a hundred bad ones
Yichen's closing line works as the conclusion for the episode. Spam is cheap now, and every platform wants to remove AI-generated spam, not because AI made it, but because it gives users nothing.
"Rather than make a hundred bad videos, make one excellent one."
Ashley tried AI for her visuals and in the end had them hand-drawn. AI can do more and more, but the standard for good hasn't moved. Taste doesn't lose value when tools get cheap. The opposite.
*This article draws from AI for Beautiful Business EP02, a series on my Chinese-language podcast.