Results Are the Marketing

I recently had separate coffee chats with two people who I think make the most honest and useful content in the Korean AI scene right now. One is Hwang Hyun-tae, CEO of SpaceY, who runs a YouTube channel called "Prickly AI" (까칠한 AI). The other is Lee Jung-min, former CSO of Smoretalk, who is part of the same Google AI Academy cohort as our team.

Hyun-tae's raw, unpolished vibe on YouTube is really appealing and hooks you right away. The topics are unique, and they dig in so deep that the insights and war stories they share are something you could never make with AI. People on our team share the channel all the time and keep saying how fun it is, so I reached out, and they gladly made time for me. Jung-min breaks down all kinds of ways to use AI in card news (short slide-style posts) that are very easy to read and understand. The quality is good enough that the head of OpenAI's Korea office has shared them.

Anyway, there's a world of difference between people who have really been through the grind in the field and people who only talk in meeting rooms or lurk on social media. You can tell what someone's got within the first few sentences. So these days, honestly, I don't really want to talk with anyone unless they're deeply into their own work and thinking hard about what's next. If on top of that they're humble and good at explaining their thinking directly, there's no way they won't do well.

Interestingly, both of them landed on the same core point. Back in January or February, maybe, but now in May, the "AI one-click show" (flashy demos where you click once and AI does everything) is a real question of responsibility. It's time to stop the fibbing and prove it with money made from AI, with results at the organizational level. I don't only hear this from the two of them. I hear it a lot in meetings. Meanwhile, plenty of people are still selling lectures for tens of thousands of won (tens of dollars) where the "hands-on practice" is making a game with Claude Code or having it tidy up your desktop.

Marketing that plays on FOMO and the sell-a-course style don't work anymore. Just look at the CEO of Ajeongdang (a Korean telecom and home services sales platform), who recently proved it with a single video of a billion won (about $700,000) landing at a time. Forget the grand narratives. Showing that you make money with AI is, in the end, the best marketing for your own skills and your service.

Beyond that, the details they shared were so useful that I wrote them up separately, and I plan to bring both of them to a Claude Bloom event soon for a panel talk.

1. Where the AI one-click show ends

Hyun-tae said this pretty firmly. Building something with vibe coding and playing around is a completely different thing from actually using it in front of a client. You can sharpen your skills and build toy projects, but the moment you meet a real client, there are many cases where you can't use those skills at all.

Closed networks are a clear example. Financial firms and brokerages can't just plug in an outside model because of security. During the PoC the environment is open, but the moment you move to the main contract, work security and internal regulations all come in at once. Most people who have this experience can't talk about it. It's partly know-how and partly confidential. Hyun-tae's conclusion was that most people going on about this topic on social media haven't actually done it.

Jung-min pointed to the same thing: a Korean AI scene where the talk keeps going with no results. No matter how much you go on about being good at AI, you get stuck at the question, "So, have you made any money?" In Korea right now, there are fewer cases than you'd think of people actually generating revenue with AI. So the critics aren't completely wrong.

2. The weight of being one person doing the work of a hundred

What Hyun-tae's team struggles with most is the higher productivity itself. When vibe coding lets one person do the work of a hundred, the weight of the projects that one or two people carry naturally goes up too.

Say an FDE (forward deployed engineer) is really good at vibe coding. Then the client asks if they can build a company-wide attendance management system. Technically, it's possible. So they take the call, and the weight changes the moment they hear that if this system throws an error, that month's payroll could be calculated wrong, which could lead to people missing their credit card payments. There's actually an FDE on the team who earned 20 million won (about $14,000) last month alone, and Hyun-tae said that person is now in tears. Because it's just too hard. Hyun-tae calls this "vibe coding blue." A feeling that didn't exist back in January or February.

What solved part of this problem was a PM with a lot of experience. On teams where a senior PM born in the 1960s was attached to the project, the FDEs' mental state was noticeably better. When the client says something like, "If this goes wrong, we could get sued," that PM shuts it down: "This is our developer. If there's an error, you fix it, right?" Hyun-tae said Palantir's FDEs hold up in a similar way. They talk loosely once a week and do only what they want to do, and that's how they get through enterprise projects. Because once Korean people start accommodating, there's no end to it.

3. Doing it alone

So Hyun-tae's observation is that one person can handle a PoC alone, but not the main contract. Technically, nothing changes. During the PoC you've already tested it company-wide, and the main contract is just pushing that to production. But in reality you want to say no and run away. Hyun-tae admitted that if they were in the FDE's position, they'd probably want to just turn it down too. Building, by yourself, a solution used across an entire enterprise comes with a weight you can't get around. That's the form of AI blue Hyun-tae is feeling these days.

Jung-min talked about the reality of doing it alone from a different angle. After leaving the company and registering as a sole proprietor, everything the CEO used to handle while Jung-min was on the team, from setting up a corporation to opening a business bank account, now has to be done alone. Development is solo, meetings are solo. Jung-min said most people aren't good at doing all that by themselves, even though they know they have to.

Even so, the direction of what Jung-min wants to build was pretty concrete. First, something that makes money right away. Second, something they end up building because they need to use it themselves. Look at what's working, like Hermes Agent or Singclo, and a lot of it blew up because the team wanted to use it so badly that they built it. Third, games. With SaaS collapsing in the software market, this direction came from the judgment that content won't go away.

One is the weight of carrying an enterprise project alone. The other is the weight of starting a company alone from scratch. The contexts are different, but both of them reach the same conclusion: doing it alone is a matter of staying power more than ability.

4. How the AX game is structured

Hyun-tae sees AX (AI transformation) as an extension of SI (system integration). Something like 100,000 SI projects get posted on Korea's government procurement site. That demand isn't going away. Only the name changes to AX; the market itself was always there. Brokerages legally can't build in-house and have to outsource. From a big company's point of view, the seat that middle-aged guys used to fill is now filled by young people. That's all it is.

He also thinks the ways AX companies make money come down to a few paths. The most basic is to keep winning projects and stack up revenue. Beyond that, some aim for an AI roll-up or an acqui-hire. In his judgment, companies that set out from day one to get acquired usually give up quickly.

Jung-min reads this moment as the right time to start a company. It starts from the observation that startups do well when the stock market is good. Expectations rise and a feeling of "I could do this too" spreads through the market. On top of that, there's the government's "Startups for Everyone" agenda and the boost AI gives you. Everything lines up this year. Jung-min added that if you've never started a company, it's just a vague expectation, but that expectation is what creates new entrants to the market.

5. Results are the marketing

How Hyun-tae got contacted by an executive at a major Korean company was interesting. Hyun-tae wrote on LinkedIn that "we've never done this," and that actually built trust. The executive said so directly: there were people shouting about enterprise AI at the time, but they seemed full of it. Someone who honestly wrote that they'd never done it, combined with being the CEO of a small team generating serious annual revenue, led the executive to decide, "We can hand this to them." That opened the door to a project at a major Korean company, and things kept rolling from there.

Hyun-tae also said they realized recently that many startups think big companies want domain knowledge, but that's not it. Startups can't have domain knowledge anyway. What big companies want is a startup's energy. Don't clumsily try to persuade them on things they don't even expect from you. Just show them a huge amount of energy. That's how you win the deal.

Jung-min's take pointed the same way. Korea needs real cases of people making money with AI. Overseas, the team that built Cal AI, the calorie-counting app, comes close. They were unknown teenagers, the service blew up, and it later sold for 50 billion won (about $35 million). Just as Yuna Kim (the Olympic champion figure skater) single-handedly grew Korea's figure skating industry, Jung-min thinks AI startups need one superstar for the ecosystem to take off. That person hasn't shown up yet. Jung-min added, quite naturally, that they hope it'll be them.

6. How each of them positions themselves

Hyun-tae described the role of their YouTube and LinkedIn as a "sedative." Executives at midsize and large companies get shaken every time a new model comes out: "Do we need to change the company because of this?" Through the work, Hyun-tae learned that someone needs to clear away that noise. For content, they screenshot the things on LinkedIn that seem like nonsense on the way home from work, film at 6 a.m., and hand it off for editing. Hyun-tae sees the real customers as the people who quietly reach out from a company email, or subscribers who just send an email saying they enjoy the videos, more than the subscriber count.

Jung-min set a rule not to promote the service directly. When results come in, they'll show them only as graphs. The goal is a brand as "someone who helps." Part of why their posts took off on LinkedIn was an algorithm quirk they stumbled onto. Post at 6:30 a.m., and since hardly anyone else posts then, it shows up at the top. Publish right when a trending keyword appears, and traffic explodes. Jung-min learned this through direct experience.

7. How they find people

Hyun-tae's criteria for hiring FDEs were unusual. Someone with startup experience, around 29 years old, and from a school that isn't too elite, just one tier below. Someone who went through a rough patch at 20, who started a company because they were good at coding and failed. People who know how hard it is to find a problem customers will pay to solve, whose pride got knocked down but whose self-respect is intact. Those are the right fit. The pay structure is a base salary, with income changing depending on which projects they land. They train as FDEs inside the team, and if they do well, they go out to enterprise clients.

The way Hyun-tae brought on a vice president followed the same logic. Deciding they needed an executive with SI industry experience, they crawled public disclosure filings and listed long-serving executives in related industries. They contacted them one by one, and the person who finally joined is the current vice president. They found a person through a system.

Jung-min talked about finding inspiration in cases rather than people. They mentioned how Sweden, a small country, still produces global software companies, and the example of Midday, a team that built a company OS for small business owners and exited to Ramp. The view was that Korea doesn't need to benchmark only Silicon Valley. Looking for hints in cases like a Nordic country of 10 million people, about a fifth of our population, expanding globally beyond the US might be more realistic.

8. On to Claude Bloom

Jung-min brought up the first Claude Bloom event first. There were some negative reviews back then, but not because of what the speakers said. It was because people had no time to talk. It was time where only the speakers did the talking. Starting with the second Claude Bloom event, we opened up small-group sessions, and Jung-min said the mood got dramatically better. Since everyone was meeting for the first time, they actually talked more honestly, and Jung-min said it made them think that people these days just miss other people.

Hyun-tae said there are a lot more AI meetups these days, but they all look alike. Within that, a gathering with information you can't get anywhere else, where people who have actually been through it in the field do the talking, could be what sets it apart. Hyun-tae also said that once enough information piles up, a community naturally burns out, the way PyCon did, and that this isn't a bad thing.

Both of them gladly accepted my invitation to a future Claude Bloom panel talk. Sometimes you need big names, but once you've built a certain level of awareness and trust, I want to bring in more voices from people in the field, the ones working outside the spotlight. Honestly, that's more fun, more useful, and it hits closer to home. Hyun-tae suggested we promote it on their YouTube too, and Jung-min said they'd start by just doing whatever they want to do. What I'm really looking forward to is seeing what happens when the two of them talk on the same stage.

Originally written in Korean on Brunch, May 11, 2026.