Two Korean Hustlers Trying to Break Into SF

Some people stick with me after a Bloom event. When that happens, I ask them for a coffee chat. That's how I ended up across the table from two people one weekday evening at a Starbucks in Samseong-dong, in Seoul. Abel Ko was a senior developer at Toss and Banksalad (two well-known Korean fintech companies) before he left to start his own company. Mark Kim has been his friend since middle school, ran an art dealing business, and then joined him. Together they founded a company called Ironact. They build AI-based marketing tools and set them up for large corporate clients, and these days they go back and forth to San Francisco for work. I first met them at the Bloom x Nebius hackathon in September. The way Abel talked when he had the mic that day stayed with me for a long time, so I asked them out for coffee.

That day I was going around interviewing participants. Someone told me Abel was the person who went viral with a post saying you could build Toss with 600 AI agents. It sounded familiar, so when I got home I looked up the post and his YouTube videos too. Toss runs on about 100 "silos," each a team made up of a PO, a designer, developers and a data analyst. He had replaced every one of those roles with an AI agent and was running silos inside his own Slack.

The agents have names. The PO is FRIDAY, the designer is Black Widow, frontend is Spider-Man, backend is Hulk, and QA is Hawkeye. They talk to each other in Slack, review each other's work, and when something gets stuck they hand it off to one another without going through him. Screen design, database design and deployment are done while he sleeps. In the morning he checks Slack, approves the merges, and spends his time on strategy. The point of the post was this: AI has always made its biggest leaps when it imitated how humans work, the way neural networks imitated the brain and LLMs imitated language. Now it's time to imitate how organizations collaborate.

SF stays vague until you go

Abel was there from the start when his company began putting together an AI team. He had watched technology improve the whole time he'd been writing code, but this time the slope suddenly got steep. He thought, this is going to be huge.

When he went abroad, his first stop wasn't San Francisco. It was New York. There he saw that the world was still a long way from AI. He looked into e-commerce and even tried running a business, but he started to worry he was drifting too far from AI. So he went to San Francisco for the first time, and last December he called his friend from there and told him to just fly out. His friend said on that call he'd come to San Francisco the next day. When they say they both move fast, they aren't exaggerating.

Mark had always had to ask developers, carefully and a little apologetically, to make small fixes for him. When he saw AI do those fixes in a single click, it made him so angry that he decided right then to go all in, and he put in 14 hours a day. According to Abel, who watched it happen, Mark picked up on the shift faster than anyone he's met, and now he builds everything himself.

Mark had spent five years dealing art. He stopped all of it: the gallery, an art center he'd opened just last year, an art data platform and an IP agency. He bet everything on AI. His reasoning was that now is when you can understand and act on the technology and the market fastest, and when you can break into a bigger market. He later wrote that over roughly ten months he put in more than 3,500 hours, seven days a week, 12 to 14 hours a day. Until last year he started each day analyzing art auction records. This year he starts it by digging through GitHub Trending, Hacker News and TrustMRR.

Early this year the two set up a Delaware corporation together, and Mark's role has kept growing since. For the first two months he led sales and marketing, designed how data and insights get delivered, and laid out the screens. The only things he built himself were frontend pieces like the landing page, onboarding and dashboard. From the third month he ran an influencer seeding automation service and built an internal operations platform that pulls market data collection, experiment logs, ad analysis and per-brand agents into one console. He also designed and developed the operations system for a distribution deal with a large overseas retail chain. From the sixth month he co-built, from scratch through delivery, a GEO platform that analyzes multiple brands under a major Korean conglomerate, an agent that turns the results into action items for each department, and a content generation engine where people can step into the conversation and make the calls.

It took much less time than he expected to take apart a codebase, understand its structure, and reliably ship services he'd built himself to big companies. In Mark's case, he had never worked as a developer, so he had less to unlearn before learning something new. He thinks that may give him a slightly different perspective.

Once you go, you get it right away. If you've never been, it just stays vague.

That was Abel's answer when I asked how they met people in San Francisco. He added that it's the same way AI feels vague to someone who has never used Claude Code.

He said people pack into AI events out of worry and fear, and most of them stop right there. Once you get your hands on it, you start thinking, "Wait, I could do this." But the mental wall in front of that moment is very high.

I learned the same thing when I started Bloom. It turned out to be less of a big deal than I thought. You just have to do it. So when people tell me they want to do something like Bloom, I tell them to start anything, YouTube or whatever. If you have your own story, there's a lot more you can do than you'd expect once you actually get moving.

You only see the ecosystem once you touch it

Before landing on what they do now, the two built a few marketing tools, tried automating influencer seeding, and on the consumer goods side even started their own AI-native brand, mass-producing ad videos and testing click-through rates. I said you don't really understand an ecosystem until you've touched it. Abel said that's how you learn, and I agreed that in the end you have to do it to know.

I'd heard they were in K-beauty, so I asked. They'd already shut it down. Once they tried it, the place where they could make an impact was too small. In that industry, being good at AI wasn't much of an edge. OEM and ODM manufacturers make the products for you, so it comes down to who spends more on branding and performance marketing.

What they do now is help brands get mentioned in AI search, the kind of marketing usually called GEO. As people start asking chatbots instead of typing into search bars, money from the already enormous search marketing market is moving there too. Once they were in it, the problem turned out to be that you can't prove the results with numbers. You can show that a brand's share of mentions went up for a particular query, but if you don't know how often people type that query, it's hard to say what the improvement led to. You can't see what's happening inside the chatbot from the outside, so you can't promise that putting in this much gets you that much back. So this kind of marketing sells on anxiety, and the numbers don't do much of the selling.

Big companies that are already good at performance marketing and have room to spare buy it first, as a strategic bet on the next generation. Brands that strictly track return on ad spend are still watching. The two landed their first client with a cold outreach, no connection at all. Someone at that client introduced them to the next big company, and that's how they reached the large client they have now.

In this field the moat is measurement data more than technology. So the companies that raise a lot of money first use it to buy large experimental samples and build estimates. Here "estimate" means statistical estimation, using a sample to calculate total query volume, and has nothing to do with an AI model. Profound, which became the first GEO unicorn this February, licensed a large amount of real AI conversation data from an outside consumer panel to make those estimates more accurate. The way the two see it, Profound started first, raised money first, and used that money to do everything it could. Data bought with capital often raises the value of a product, and once in a while, though it's getting harder, that becomes a moat.

They compared AI search marketing to a vitamin and influencer seeding to a painkiller. One is nice to have. The other you go looking for right now, because it hurts, and you don't haggle over the price. As it happened, what I'd heard that morning at a consumer goods company was on the painkiller side. Their influencer marketing team is the largest team in the company, and its biggest pain is the emotional drain of finding influencers, negotiating with them and dealing with people. I asked whether AI tools that automate seeding could handle that. The two had built one, and they said it can go as far as DMing hundreds of influencers at once.

But every brand wants a different kind of person, so the criteria never get standardized and there are endless exceptions. You're also competing on cost with agencies that throw people at the problem, so even if you're a bit more efficient, the margins are thin. Most of all, brand owners trust influencers picked by a human more.

The hardest thing to automate is human taste.

After weighing the cost of handling all those exceptions, the two changed their minds. Going into someone else's company and cutting work that's already set in stone from 100 down to 80 is slow. Using the same technology to build several brands of their own and go from 0 to 1 is much faster. Inside a company you can't cut people right away, and just getting access to one dataset can take three or six months of back-and-forth requests. What matters is building anything faster than everyone else, like running with swim fins on. They didn't start out with this view either. It came from doing the work, and they said their hypotheses are still changing.

The world is still far from AI

For a while the two stayed holed up in the office, working like monks training in the mountains. When they came out and talked to people, the world turned out to be much further behind than they'd thought. Step even a little outside their bubble and the share of people who don't use AI jumps. That held for ordinary people, and also for developers busy with their day jobs.

Abel had seen the same thing in New York. The two said even marketing teams with big ad budgets don't know which image model to use. Whoever closes that gap at the point where consumers meet the product can grow fast, they said, and that's how Canva grew.

SF networking = hacker houses

In San Francisco, networking runs through hacker houses. The hacker houses are well connected to each other. During the day people gather at one house to work together, and in the evening they go to Luma events at places like Frontier Tower. You're drinking and talking there, someone says, come work at our hacker house, and you just move over. That's how one hour of networking turns into eight. The two decided that going to the AI events listed on Luma and working all day in hacker houses was enough to make a real impact.

There are so many events that organizers serve dinner at all of them to draw people in. You can eat some pizza, talk with one or two people and leave, and the food is pretty good. You meet investors naturally at these things, and there are lots of YC founders in the hacker houses. During SF Tech Week earlier this month there were reportedly more than 1,700 events. When you hear about cocktail parties on an aircraft carrier and events in rented-out art museums, the phrase "community event" takes on a different weight.

The friends you make introduce you to more friends, and when you get invited to someone's birthday party, that's when you're really in. From then on you're friends, and the business connection is secondary.

The biggest thing the two got out of San Francisco was information that doesn't come through in writing. Join a morning meetup on Luma, like a running crew, and someone in an OpenAI T-shirt strikes up a conversation and tells you they're a Voice AI researcher. That's how you hear about companies a month before they blow up.

There's a temperature there that you can only feel on your skin. It never comes through in text on X.

Mark later added that he skipped a lot of classes in college, but during the review week right before finals he gathered everything with intense focus: the intonation professors unconsciously stressed when going over sections that would be on the exam, how long they lingered, even their facial expressions. That helped him a lot through his second year, and in his third year he ended up selling those notes to classmates. In the industry where he spent so many years, signals like that are everywhere between sellers and buyers, and plenty of people get pulled into the market by the fun of reading them.

That's the reason I do Bloom, too. Things said offline get refined again and again before they show up on X and LinkedIn. The lag has gotten much shorter lately, but knowing first still makes a big difference. In San Francisco, at the center of tech, even more so.

The two also think real conversation happens with 20 or 30 people, not 100. Mark felt this firsthand running a small community for several years, and he confirmed it again recently when he hosted a small salon on marketing and AX (AI transformation). When around 20 people came, everyone got to talk, and he said people clearly loosen up in a cultural space. That's also why I set up small group meetups at the end of every Bloom event.

Up close, everyone's about the same

Wrapping up our conversation, Mark said the reason we need offline comes down to one question: "Am I using AI well enough?" Your feed is tuned to you, and most of what gets posted is the part people chose to show. So you can't really answer that question by browsing inside an algorithm. How people who do AI for a living actually structure their products, and how they layer and manage their data, comes through much more raw when you meet them, hear them out and look at their screens. The ability he ranks highest for anyone taking on a new market is staying in a state where you understand, at a deep level, the market and the technology that keep shifting and how the whole structure fits together.

The other thing the two took from San Francisco was confidence, and it came from looking at screens like that. They'd sit with YC founders and look at what kind of work they were doing at which layer. Whatever they couldn't figure out on the spot, they'd pull apart later. It turned out to be less impressive than they'd assumed. Everyone's about the same. But then a team whose screen they'd looked at posts on LinkedIn two weeks later that it raised $7 million. That makes you want to go to the bigger market.

What separates them is the market and the capital.

People who've already made it and have a lot of money keep investing because nobody knows who will hit. When someone does hit, they invest again. That loop keeps turning. VCs are afraid of missing good opportunities, so they can't leave San Francisco. Not everyone grinds like crazy, either. Teams that work at a surprisingly relaxed pace become unicorns, and the few teams that really grind build companies much bigger than that.

One of them brought up the story of the elephant tied to a rope, in the same vein. If an elephant doesn't know it's an elephant, it stays tied to one rope. But in a big market, even a monkey thinks it's an elephant, so it makes it at least as far as a gorilla. Where you are decides how big you see yourself.

For capital, London comes after San Francisco and New York, and most big tech companies and AI labs have their second base in London. People pile into London too, so much that there's a saying: if you don't have a US visa, go to London for now. The number of events on Luma follows the same order. If you do fashion you go to Paris. If you're a musical theater actor you go to Broadway. If you do AI, it's San Francisco. That's why I want to go to San Francisco too. If you make it in the holy land of tech, the global part follows.

I've written up stories from people who went to San Francisco before me in two other posts (in Korean on Brunch): "Bloom, Get to San Francisco Now!" and "SF = Spotting the Real Ones Among 30,000 Elon Musk Wannabes."

Moats belong to people who've stopped

A moat is what you obsess over once you've decided you're done getting better.

Abel said that if you keep running, it doesn't matter much how fast competitors are catching up. He'd rather open up the technology and grow the market than hold on to it. Like Tesla opening its patents to grow the EV market: without the shield of a patent, you have no choice but to keep working, and that kind of culture moves fastest.

He doesn't see the world as zero-sum. Instead of splitting one loaf of bread, you can bake two or three. There are still lots of hard problems nobody has solved, like getting agents to handle more than 200 back-and-forth exchanges on their own, and he wants to solve those.

He thinks about money the same way. He said even people who made it big in Silicon Valley buy the yacht, the NBA team and the watches, and then come back saying they just like building products. Seen that way, him doing this now and him doing this after becoming worth 100 trillion won (around $70 billion) would look the same from the outside. So he wants to do work with impact, work that helps humanity.

Mark said a moat is closer to an illusion. Pouring money into measurement data doesn't mean "don't follow us." You do it because you need it to build a better product. That's different from kicking away the ladder, like bidding up ad prices so latecomers can't get in at all. Still, some people do start kicking away the ladder once they see the market has a ceiling. Just as there are investors like Warren Buffett and investors like George Soros, he said, it comes down to temperament.

I'll go first and break the ground

One of them said you can get anywhere on Earth in a day, so he takes planes the way you take the subway. The two also said they won't take the path of becoming an agency that wins a few outsourcing contracts and makes decent money. That path has no next step, and you never get to plant a flag.

Abel brought up the story of Chung Ju-yung, the founder of Hyundai, who supposedly walked into a British bank holding a banknote with a turtle ship on it (the armored warship from Korea's 16th-century navy) and borrowed the money to build a shipyard. An idea like that had no data behind it, he said, so AI could never teach it to you. Chung Ju-yung and Lee Byung-chul, the founder of Samsung, both saw their businesses wiped out by war and still got back up and built companies. He doesn't think he'll face anything harder than that, so he has to do it too.

When I held a Bloom event in Singapore, I met a friend from Europe. In their home country and in Singapore, their network was limited, so it was hard to get anything going in AI. They said they were going to San Francisco no matter what, and once there they kept pushing and pushing until they even got into an Elon Musk event without a ticket. Watching that friend, I think you really do have to go and throw yourself into it. Going there and posting your own photos beats scrolling Instagram in Korea and thinking it looks cool. You keep building your network, interviewing people, and climbing.

Abel said there are Koreans who went to the US early and are doing well, but when he meets them, many have a lot they want to do and would come over with just a little push, yet haven't jumped in. So he'll go first and break the ground. Later, if someone is on the fence, he wants to build a community that gives them a room and says, come try it. He doesn't even like K-pop much, but he said to get people together he'd probably have to start with a K-pop party. I chimed in that just serving tteokbokki would bring everyone.

On Reddit, and in Seoul

As a way to put yourself out there without moving, the two recommended Reddit. If you have a tech product, you need to post in overseas communities where people try all kinds of tools, and posts on Reddit get cited a lot in AI answers. An acquaintance Mark has been close with since college built an API crawler service as a side project. They prepared GEO and SEO carefully with Reddit at the center, organic traffic took off in North America, and within six months it was bringing in $30,000 a month in recurring revenue.

One of them said your own channel is money. Now that anyone can build things easily, the people others trust and follow matter more than the technology. When there's too much to choose from, people buy from someone they want to trust, the way influencers called wanghong drive sales in Chinese e-commerce. If you have a channel, even something odd you put out gets seen by tens of thousands of people, so you don't lose.

The best part of doing Bloom is that I keep finding underdogs I'd never have seen otherwise. Finding people like that and getting to know them is fun and I learn a lot, so I ask for coffee chats whenever I get the chance. This meeting happened because Abel caught my eye at the Bloom hackathon.

After Abel finished speaking that day, people came up to him with all kinds of questions, and he said it gave him the courage to believe he had insights worth sharing too. In the end, he said, connecting people is the biggest thing, San Francisco is good at it, and Bloom is doing that for AI right now. About Bloom, they said that when overseas tech companies look for a big AI community in Asia, AI will recommend Bloom. They also said the developer conferences that used to run big on corporate sponsorship could be what Bloom does for the next generation. When they said they like that non-developers can come in comfortably and learn from developers, it came with an explanation: most developers like teaching, and open-source culture is in their bones.

In San Francisco, one meetup turns into eight hours. In Seoul, people still scatter as soon as an event ends. Mark said he hopes Seoul's Luma listings will have more events led by communities instead of companies. I want to build that scene in Seoul too, stretching that one hour a little longer each time.

Originally written in Korean on Brunch, October 27, 2026.