Playco × Bloom: A Fireside Chat with Teddy Cross

I hosted a Claude Bloom event with Teddy Cross, a friend I've known for 11 years. We first met at Startup Grind, a startup event that, unusually for Korea, ran entirely in English. Back then there weren't nearly as many foreigners in Korea as there are now, so meeting him was a nice surprise and also felt a little strange. Later, during my military service, I happened to meet a developer in my unit, which got me more interested in tech startups. After I was discharged I took an internship at a startup called Buzzvil, and found out Teddy was working there as a software engineer. Funny how these things work out.

We stayed in touch and met up now and then to catch up. This time I invited Teddy as a speaker, and we did a fireside chat together in front of a hall packed with about 120 people. Afterward, talking one on one, we both realized that somewhere along the way each of us had become someone who stands on a stage. We laughed about how strange that felt.

I don't think it was only the two of us who felt something. Insights from people working outside Korea are new in themselves, and the range of perspective is different. Personally, I want to run at least one English session every month. It's less for the foreigners living in Korea and more for Koreans. I feel we live in an environment so cut off by language, so Galápagos-like, that we don't even notice we're stuck at the bottom of a well. I really want to break that open a little. With K-pop and other things making Korea more global, more and more foreigners are coming here, so I want to hold these English gatherings as often as I can and connect people. It's a big push for me and for my team too.

Anyway, Teddy co-founded a global unicorn valued at about 1.4 trillion won (roughly $1 billion), with investors including Sequoia Capital. Total funding raised comes to 230 billion won. In this post I'm going to write up the session in almost excessive detail.

Insights don't happen at conferences

On Thursday, June 11, we held Claude Bloom × EF × Playco × Tiro at the 6th-floor event hall of D.CAMP in Seolleung, Seoul. This time our guest was a Silicon Valley unicorn founder.

About half the people there were first-timers. The best part of a Claude Bloom event is always the small-group sessions. I believe more comes out of talking with the person next to you than out of the talk on stage. So I tell first-timers in advance to share a lot during the small groups.

Claude Bloom is a community where people going through AI blue think things through together and share what they've learned, so they can bloom. (I wrote about this in Claude Blue and From Claude Blue to Bloom.) A lot of people at this event were feeling AI blue. Someone I met before the event, who used to work at Google, said they were feeling it badly. Everyone is going through such a confusing time that some people can't even tell whether they're feeling AI blue. Since this was a room of people who think alike, I hoped we could talk honestly.

The topic was a conversation with a Silicon Valley unicorn founder, but what mattered more was overseas AI trends you won't even find on X. I want to bring new content to every event. Everyone already sees Korean content on YouTube, LinkedIn, and Facebook, so what's hard to get is what's happening abroad. I can't always fly people in, so I wait for people who run businesses overseas to come to Korea and grab them while they're here.

The speaker was my American friend Teddy Cross. When we met, Teddy was in his late teens and I was in my mid-twenties. I had just gotten out of the military and was interning at the company where he worked, so it's been more than ten years. We've talked about a lot over the years, but this was the first time we'd sat across from each other in a professional setting, which felt special.

Teddy has been a big influence on Claude Bloom. He once told me that these days, whether you're in Silicon Valley or Seoul, real insight doesn't come from conference halls. Content was oversaturated even before AI, and large-scale networking doesn't mean much. Real insight comes from rough conversations before anything goes online: friends talking at a small coffee shop, or eating and having a few drinks at a friend's place. I get unexpected insights myself every weekend having coffee with friends and catching up. That one comment from him is why I've kept running offline events.

Teddy's story is unusual. He was homeschooled, grew up in a small city in Tennessee, and moved to Silicon Valley alone at 15. The reason was unusual too. He posted a game he'd made online, it went viral on X and hit number one on Hacker News, the forum run by YC. Someone in Silicon Valley saw it and sent him a cold email, and at 15 he went to San Francisco by himself. Later he came to Korea and also spent time in Tokyo. He kept growing through hard times, and our two-hour prep conversation was packed with good material. As the event description said, he's a founder of Playco, which raised a total of about 240 billion won from Sequoia and others through its Series A and built a company valued at around 1 trillion won.

AI blue is also a good topic to put to someone from abroad. It isn't only a Korean thing. If anything, people overseas are more exhausted by AI productivity and more burned out, so it's worth continuing to discuss.

Our partner for the day was EF, the global education company. EF sends people abroad for language study and also brings people into Korea. That aim lines up with Claude's global direction, so we teamed up, and EF sponsored the swag people took home. There was another point of overlap: the insights you need in the AI era are limited online, so the ability to go abroad and dig them up, and to dig them up offline, matters more. The fireside chat was in English, and Tiro showed a Korean translation live with its real-time interpretation app.

1. Why EF was here

Jaekyung Oh, country manager of EF Korea, spoke first. He explained why an education company was suddenly speaking at an AI community event.

EF is a Swedish education company that has connected people around the world through language education and international experience for 60 years. It first came to Korea during the Seoul Olympics to teach English to police officers, students, and volunteers, and has operated here ever since.

He said they're wrestling with a question. As a foreign-language education company, when AI translates and interprets English for you, will people still go learn a foreign language? How does this business survive? What stood out to me in that context was EF's AI division. It's an online learning tool they call Efekta, and that division made TIME's list of the 100 most influential companies in the world, alongside NVIDIA and Anthropic.

His point was that EF didn't get on that list by suddenly building some technology in the AI era. It was recognized for bringing 60 years of language education and international experience into AI. So rather than talk about AI technology, he wanted to talk about what abilities help you survive in the AI era and what language education and international experience will mean.

His last point was that, going forward, everyone starts on the same line when it comes to language. Speaking some English won't be an edge anymore. Language will be a tool for meeting people and creating new opportunities through them, and that's why EF sponsored the event. Once language is only a tool and no longer a barrier in the AI era, the experience of going abroad and breathing the same air as people there becomes what matters.

2. From a small town in Tennessee to Silicon Valley

Teddy spoke in English. He speaks some Korean, but he's American and more comfortable in English, and with the live translation on screen we didn't have to worry.

Teddy is from the US and lived in Silicon Valley before moving to Seoul. For the past ten years he's spent about 70 percent of his time in Korea and the rest in Tokyo and San Francisco. He's a co-founder of Playco. They make games and now build AI to make games, and he said that for the past year or two, AI is the only thing on his mind when he wakes up and when he goes to bed.

He said his childhood was extremely boring. He's from Chattanooga, Tennessee. It's technically a city, but not by Korean standards: a mile from his house there were farms. I lived in Michigan, so I got it. Since he was bored, he read a lot, mostly science fiction and Wikipedia. He said Wikipedia saved his life.

As a kid he wanted to be an inventor. He didn't really know the word entrepreneur. He just wanted to make things, so he built with Legos, did science experiments, loved Duplo and Lincoln Logs, and built with anything he could get his hands on. That moved naturally into coding, and he loved building games like Minecraft. "Inventor" is basically what we'd now call an entrepreneur, so I think he was a born entrepreneur and inventor.

The roots of his story go back to a game he made at 14. Flash games were big then, and Teddy was hooked on tower defense. Steve Jobs had killed Flash, but it wasn't completely dead yet.

In 2010 there was a contest sponsored by Microsoft. The theme was building a web app in HTML5. Web apps were brand new in 2010, nobody was making mobile web apps, and everything was desktop. Lots of people still used Firefox, and Chrome was new. Everyone else was building web apps, but Teddy had never seen anyone make a game with web technology, so he decided to make one. He spent about ten days rebuilding a favorite game of his, one that ran in Flash on desktop, using web tech, and submitted it.

He didn't win. But he wanted a leaderboard in the game, and the contest rules didn't allow a server. While he was stuck on that, he thought of Twitter. Twitter wasn't that big outside Silicon Valley then, but it was catching on there, and it had an open API anyone could use to post and search. So he made a hashtag for the game and added a leaderboard component. When you lost, your score got posted to Twitter, and it showed up on the leaderboard. It went viral the moment it launched. People in the contest posted their scores, their friends saw them, played, and posted too. Eventually someone posted it to Hacker News, and since nobody had made a game with web technology, it went straight to number one.

He was aiming for the contest, didn't win the contest, and went viral with the business crowd instead. It's a good example of how life actually goes.

3. Silicon Valley, and then Korea

About a year after the game hit number one on Hacker News, something a little crazy happened. A CEO in Silicon Valley who had seen the game the year before reached out. His company was building web games with web technology, and he sent over a TechCrunch article about their fundraise and asked to get on a call. It came as a GitHub message, not an email. GitHub had messaging back then and later removed it. Skype came up during the call too, and Teddy said he still doesn't understand why Skype lost to Zoom.

When they offered him a job on the call, Teddy said, "I'm 15." He did a few more interviews with the co-founders, they talked to his parents, and lawyers on both sides worked out a way for a 15-year-old to work legally in California. Two weeks later he flew to Silicon Valley and never looked back.

There was something he only found out years later. His mom went with him and met with the startup accelerator, the CEO, and the co-founders, asking things like "What are you going to do with my son?" and "Who are you?" She was doing due diligence to make sure they weren't kidnappers. Teddy had no idea at the time. He said he met plenty of high school dropouts in Silicon Valley, but never anyone else who had never gone to school at all. So he was an odd one even in Silicon Valley.

That first experience shaped his future. He followed Hacker News closely and is still addicted to it. Silicon Valley was his dream, but he never expected it to happen so fast, and when he got there it was even better than he imagined. He met people he had a lot in common with, went to the Google campus for lunch, and walked down the street past startups he'd only heard about in the news. He decided he had to stay, and he did. He's been in startups ever since and made a lot of founder friends, and that shaped every part of his life. The first thing he did when he came to Korea was meet me at Startup Grind. He liked living in Korea, but he wanted to know where Silicon Valley was here and how to find it.

He also talked about daily life there. He lived on California Avenue in Palo Alto, the heart of Silicon Valley. Housing was so scarce that two weeks before he moved he still had no place, and he ended up living with four Stanford PhD students. He paid $800 a month for a room the size of a closet, which by Silicon Valley standards was a great deal. The company was part of StartX, Stanford's startup accelerator, a place a bit like D.CAMP, so he was surrounded by entrepreneurs working on all kinds of things. People are shaped a lot by who's around them, and this showed how important it is to be surrounded by people who think like you.

If going from Tennessee to Silicon Valley was his first big shift, crossing the Pacific to Korea was the second. Back then people thought of the UK, Europe, Japan, or China, and Korea was less well known. Coming here was an intentional accident. He did an Asia trip with friends from Silicon Valley. In Silicon Valley nobody takes time off except in December, which is a dead month when everyone goes home. He traveled around Seoul and Taipei for a few weeks with Korean American and Taiwanese American friends and spent Christmas 2013 in Seoul. Within a week he thought, "This is the future. I need to live here."

He compared the cities. For someone from Tennessee, arriving in Seoul was wild. He'd never seen that kind of round-the-clock energy, and as an American he was amazed that Samsung had built the apartment complex where his Airbnb was. He didn't know Samsung built buildings. Seoul was futuristic and modern but still felt familiar and comfortable. Taipei felt much more foreign, and he could picture himself living in Seoul. It helped that English didn't work at all in Taipei and worked well enough in Seoul. He made up his mind at Christmas 2013 and moved in March or April 2014, twelve years ago.

He also talked about working at a Korean company. It was Buzzvil, and going from an American company to a Korean one was a shock. In Silicon Valley a lot was theoretical. Back then plenty of startups had zero users. Raising money with no users, building lots of product, pivoting several times, and eventually finding users was very common. Korea's funding environment would never allow that. At Buzzvil he learned how to scale large production engineering systems in an environment with lots of users, and felt that limits and pressure make people grow. Buzzvil was very forward-looking then. Everyone used English names, and being part of the future of building startups in Seoul was a great experience, he said.

We moved on to the global mindset. I studied and worked in the US and came back to Korea, and here I hear a lot that having a global mindset is important. In the internet era we assume everyone is connected internationally, but once you live here you become more Korean and more local. Other countries are the same. Teddy said Silicon Valley sometimes lacks a global mindset too. It's a giant bubble that sometimes forgets people exist outside it, and that creates its own problems. A global mindset is good for both sides, whether you're in Korea, Japan, or Silicon Valley, and the core of it is empathy and nuance. The more global you are, the more empathy you can have for different countries, cultures, cities, and people. If you're building a product for the whole world, you create the most value when you're truly global.

I asked how to get out of your local bubble. Teddy said the internet is great. He was in the same position about Silicon Valley when he lived in Tennessee; Silicon Valley is as far from Tennessee as it is from Seoul. There are lots of resources on Twitter, blogs, podcasts, and YouTube, and if you follow the most trustworthy, most current signals from those sources you can get far. But in the end the best thing is to go and see somewhere different in person. He joked that going to Silicon Valley and meeting Koreans is the last thing you should do. Immerse yourself in a different environment.

4. How Playco came together

I asked about Playco's background. It was founded by Justin Waldron, who co-founded Zynga; Michael Carter, who invented WebSockets; and Takeshi Otsuka, who built businesses in Japan. They were all friends and colleagues. They had worked together or near each other, knew each other, had all done web games in different ways, and were all in Tokyo at the time. Michael and Teddy met in Silicon Valley, Michael and Justin had met many times at conferences, and Take and Justin knew the same people in Tokyo. Since they were all living in Tokyo they came together naturally. They had shared experience and saw the same opportunity, so the idea was to combine forces and build something to go after it.

WebSockets is the technology behind all real-time communication on the web. Slack, ChatGPT, anything that works in real time uses the WebSockets protocol, and Michael wrote the first spec. He's spent decades in web technology and open source. Zynga is the gaming giant behind FarmVille and a long list of other games, and Justin was a co-founder there.

I asked why they met in Tokyo. Japan is the second-largest game market after the US. China is its own thing, so the US is first and Japan is second. Tokyo is the gateway to Asia, very international in some ways, and close to San Francisco and Los Angeles.

I happened to have been in Tokyo the week before, so we kept going on that. I knew it was international but didn't realize how much. Shibuya is half tourists, and it was diverse enough to feel like an American city. Some Japanese companies use English as their official language, like Rakuten and Mercari. They're Japanese companies whose main market is Japan, and their working language is English. Teddy has hired several engineers in Japan, and none of them were Japanese. A friend of mine who's a PM at Mercari said their team has 20 people and only two are Japanese; the other 18 are from other countries. I've heard that Japan doesn't have many IT people who grew up or were educated there, so companies have no choice but to bring in talent from India, Australia, the US, and elsewhere. People think Japan and Korea are similar, but the environment makes them different in places.

5. Hiring in the AI era

Teddy is hiring from anywhere in the world right now, so he's been thinking about hiring a lot. The role he's looking for is product engineer.

He said hiring has completely changed in the last six months. You still want good people, people who are easy to work with, senior people. That hasn't changed. But testing engineers on coding is over, because nobody codes anymore. Goodbye, coding interviews. You might still look at architecture.

He had strong views on management too. He sees hands-off management, where you step back and watch, as completely over. Some organizations used to have a reason for managers like that, but since AI, outside of extreme cases, that need is gone. Now everyone builds. Engineering managers code with AI. Big Tech is cutting middle managers. Teddy himself took a few years off from coding, and now gets a lot done coding 15 minutes a day with AI. So he builds smaller teams, and each person on them does far more, in both output and scope. The people who are good at this are often different from the people who did well before AI.

Where engineers spend their time has changed. Coding is now a very small part of the job. Now the time goes into defining the product, writing specs, doing research, talking to customers, building, shipping, and testing. So he's looking for people who can build something from start to finish, more than engineers who only code.

I asked how you test for that. With coding you can see what's right and wrong, but building, shipping, and finding customers is hard to see. Teddy laughed and said if you can answer that, come work with him. What he means is taste, especially product taste, and it's really hard to test whether someone has good taste. One way is to look at what they've built before, ideally what they've built alone: personal projects and a portfolio. A good portfolio is a pretty good signal of good taste, but some people have good taste without a portfolio, and that's the biggest challenge.

Before AI you had to work with designers, marketers, and PMs, so you had to be at a company to build anything. Now you build a product alone with AI. Teddy said having a portfolio wasn't that hard even before AI, and he had one himself. Now you can build a website with AI in five minutes, so there's no excuse for not having one. Many companies also give a take-home assignment to build something related to the business. If it looks good and feels good, that's a good signal, but it isn't everything.

We talked about how this change won't stay inside tech. Teenagers in school are still measured by tests like the SAT, which is outdated and far from a tech company interview. It's affecting non-tech industries too, and it will take a few years to reach academia, but we're looking at the future right now.

That led to retention. Now that you can build good products with AI, keeping top talent is harder: a small team builds something great, and that team might leave. Teddy said two-person unicorns already exist, so the first solo unicorn is coming soon. What one person can do is unprecedented. But the job market isn't good, and building your own thing takes stress, time, and effort. That's great for people who want to take the shot, but many people won't. Teddy's answer for keeping those people is to make building really interesting and fun. Playco is doing interesting work, and he wants employees to know Playco is the best place in the world to stay at the frontier of AI. On top of the job, you get an education and experience you can't get anywhere else, and people care about that. Being surrounded by the best people, and their personalities and teamwork, is what holds a team together. There aren't many places outside Silicon Valley where you can learn AI this fast, and Playco offers that. It isn't easy, he said, and you have to work really hard to make it possible.

6. Living as an AI-native company

Playco's team is spread across the world, so I asked how full remote works and whether AI has sped it up. Teddy said AI hasn't changed the remote side much. They keep flexible schedules and group teams by time zone, so Asia works with Europe and Asia works with the US, and they avoid calls where three time zones overlap. They do a lot of synchronous, human-to-human communication, and AI makes the rest easy. Company-wide tools like Fireflies have been useful, and agents do work. They've been remote since before COVID. The number one way to connect people remotely, he said, is turning your camera on. Not always, but when most people have their camera on most of the time, it's worth a lot. They deliberately have short daily meetings, not long ones, so that nobody in the company goes a day without a meeting. Human connection matters, so they create a lot of touchpoints.

We moved on to AI-native organizations. In short interviews before the event, a lot of people wanted to know how AI trends in Korea differ from elsewhere. Teddy again pointed out that Silicon Valley is a bubble. It's the only place where AI is in the air everywhere: your Uber driver and your friends talk about AI, every billboard is AI, and the rest of the world isn't like that. Korea is fairly far ahead, though, and one of ChatGPT's top countries. "AI-native" is a term used in the US too, and getting an existing company to become AI-native is a big problem and a big topic. In some ways it was really hard for Playco, and most companies are failing at it.

He described daily life at an AI-native company. Every morning starts with the question "How can we be more AI-native?" and never stopping that question is the baseline. One way to define it is that everything is AI-first. When a problem, task, or routine comes up, you go to AI first. When everyone does that consistently, it makes a big difference and puts you at the frontier. A lot of it is just hard work. So that everyone stays current on the latest models and trends, they watch X, keep several AI channels, and constantly share the latest news and updates. It takes a lot of effort to stay AI-first and keep AI from becoming an afterthought.

We also talked about token maxxing. A few weeks earlier there was a lot of talk that the more tokens you use, the more professional you are, but as AI costs have approached the cost of a person, the mood has shifted. Teddy said Big Tech keeping a leaderboard of the biggest token users is a terrible idea, because all people will do is set up something that burns tokens while doing nothing useful so they can get to the top. They do it to get promoted or avoid getting cut, and it was never a good idea. That said, for years Playco has made sure nobody has to worry about AI subscriptions, tools, or token costs. Art, tokens, new tools: people get what's effectively an unlimited budget with almost no questions, and if expensing is a hassle they get a virtual credit card to spend directly. Making the friction zero for switching to AI, trying new tools, and using AI is one of the most important ways to become an AI-first company, he said.

I summed it up this way: we're all in the process of becoming AI companies, so maybe the top 1 percent or fewer can call themselves 100 percent AI-native. The older the company, the more systems, protocols, and non-tech staff it has, and the harder the transition.

I asked how the company changed before and after AI and where the bottlenecks were. Teddy said he heard Andrej Karpathy, the OpenAI co-founder, say on a podcast that even he sometimes feels behind. If Karpathy feels behind, Teddy said, he's really behind, but that's completely normal. You accept it and keep going. At Playco it started with the art team. The art team has been AI-native for three years, expanding gradually from backgrounds to characters, animation, and video. The engineering team took much longer. AI wasn't good enough to beat them, but they knew it would be someday. Every week a new model came out, they tried it, it got better, and when it finally got good enough, they switched. ChatGPT has only been out for three years, and at the start there was a lot of news telling people to stay away from AI because of hallucinations and risk. Teddy said it's still full of hallucinations, but you only see them if you look closely. It isn't 100 percent, but it keeps getting better, and it depends on how you use it.

We talked about articles saying AI use isn't turning into revenue: everyone uses AI every day, but revenue isn't growing. Teddy said he read the same articles and can't agree. If that's your experience, he said flatly, you're doing it wrong. From what he's seen, the art team is producing 20 times more content, or 10 times more content at twice the quality, and the engineering team is far more productive too. It's not only engineering and art. The head of marketing vibe codes, artists vibe code, everyone at the company vibe codes, and they're opening up workflows on their own that were impossible six months ago. What they spend on AI is tiny compared to the value it creates, he said.

7. Utopia or Skynet

When we met a year or two earlier, Teddy said that in five years we'd see either utopia or Skynet. Now there are two or three years left. Fable had come out the day before and everyone was trying it, but Teddy said that, honestly, he prefers the GPT models.

I asked how he felt the first time he saw ChatGPT. It was different from the metaverse or blockchain, he said. On one hand it was pretty bad. On the other hand it answered anything you asked with some accuracy and wrote a convincing poem in the style of Eminem. He thought: if this is the alpha and it's already this good, what will the beta be like, or version 1, or two years from now? If a technology this new can do this much, everything will change completely when it gets really good. He picked up AI at that moment and never put it down, but he didn't know how fast or how dramatically it would happen. Around GPT-3.5, when image models like GPT Image 1 arrived and agents started to show up, he saw it was accelerating, and understood more. That was the context for talking about which future we'll end up in: Ultron, Skynet, or the Matrix.

I asked what odds he'd put on positive versus negative. He said he doesn't know. Many people in Silicon Valley are worried and trying to steer it toward the positive, but at the same time there's an arms race over who has the better model. Two years ago people were saying slow down and regulate, but once the money came in that was all thrown out and it became "as fast as possible." Chinese and American models are fighting for first place while pretending to be safe, and in the end it's about who controls AGI first. He hopes it won't be a disaster, but it could go either way.

I asked how he defines AGI. Teddy said it might be consciousness, but it's definitely self-improvement. Once we reach the point where long-term memory, long-term learning, and self-improvement are solved, self-improvement will come from long-term memory and learning. After that it's a matter of giving it the right tools and systems, and today's AI doesn't have the tools to become AGI. It needs to be multimodal, with vision and text in the same model, and that isn't so far off, he said.

What struck him is that people don't know this is coming. The people in this room know, but 99.99 percent don't. They say "AI is bad," but AI is inevitable. He was amazed that the people here are the top 5 percent who are truly passionate about AI, and that there's a public that isn't very interested in it. The public uses ChatGPT, but doesn't think about the future of AI as deeply as we do.

8. The one thing he won't hand to AI: writing

The funny thing about talking with Teddy is that he uses AI for everything except writing. He never uses AI to write. I asked if he has a philosophy about it.

Teddy joked that the last thing we'll have as humans is grammar mistakes. AI will have everything else, and we'll have grammar mistakes. More seriously, he said it's about authenticity. Anyone who's done marketing for a while knows authenticity is king. UGC, TikTok, YouTube, every channel that reaches people is about authenticity, and the moment you let AI speak for you, authenticity is completely destroyed. It isn't you anymore. If you review it carefully and polish it so it doesn't sound like AI, you've basically written it yourself. If you need a formal speech, you can have AI write it and fix the mistakes, but that's still a way of doing your own writing. Grammar matters, but it isn't a strict requirement. When you talk with friends, teammates, and customers, authenticity matters most, and AI takes that away.

I took it to mean writing is a mirror of yourself. What you write is who you are. Teddy added that the way we communicate with AI is all writing, and even when you speak, it gets transcribed into text. Writing is the highest-fidelity way humans express themselves, so if you lose that ability, you lose part of what makes you human. Teddy keeps AI out of his writing so he can stay as human as possible. Writing aside, he hands almost everything to AI.

9. How to stay connected to the frontier

Finally, I asked for advice for people in Korea and around the world on connecting to the frontier of innovation. Teddy said that, as even Andrej Karpathy shows, everyone is in the same boat more than you think. So use the best of the same tools strategically. Unless you're someone like Elon Musk, you can't spend 24 hours on Twitter, so you have to be strategic. Podcasts are an example. They're hard to fit into a schedule, but when one goes really viral or has an interesting guest, that's your chance, because you get to hear directly from people who aren't just at the frontier but are making it. The good thing about the internet is that it gives everyone the same resources wherever they are, and you have to actively take them and do something. Listening isn't enough. You have to build something yourself, share it, and move forward from there. Online or offline, ideally both: immerse yourself, build something, and share it with people, and you can get pretty far. That message summed up the whole conversation.

Before moving to the small groups, we took two audience questions. First I asked Teddy to wrap up in Korean. He laughed and said he wasn't ready, he was fully in English mode, and he hadn't thought in Korean for days. So I asked the question I figured many people were curious about, how he raised money from Sequoia, and he answered in English first. There are three co-founders, and Michael and Justin are legendary figures in Silicon Valley fundraising, so they handled all of that. Having great co-founders is really important, he said. He never could have done it alone and is lucky to have the team.

In Korean, he talked about motivation. He's motivated by AI and wakes up every morning excited about it, because it's the future. We're all building the future together. Never in human history has anything moved this fast and this globally while giving every individual the chance to build something or take part in shaping the future, and being part of that is his biggest motivation every day. Asked about Japanese culture, he said honestly he isn't that interested and only likes the food, and that he likes Korea much more, which got a laugh.

10. What came out of the small groups

After the fireside chat we moved into small groups. The point is for each group to share what it discussed with everyone else. I was most curious about this part myself, so I asked each group leader for their name, their affiliation, who was in their group, and what they talked about. There were two questions: what AI trends are you interested in these days, and how do you collect that information and make it your own.

The group led by someone who develops AI concepts at an international organization had four people, all running a business or interested in one. For trends they picked an AI agent analytics tool YC launched on June 6. It can analyze Codex, Cursor, and Claude: you type one line in the terminal, it figures out how you use agents, analyzes your repository, gives you a report, and lets you talk with the agent. They also talked about knowledge graphs, the second brain making a comeback, more interest in memory management because of context overhead, and a video that made them think about AI ethics. On collecting information, the problem is that there's too much, so the challenge is filtering for what's meaningful, deep, and actually useful to you, and making it your own. The leader said they use a primitive method. Humans differ from AI in that we keep thinking, and the more you use AI the more you risk losing the ability to think, so it's important to write down what you understood and share it with people.

A group with people from marketing, SK Hynix, and TikTok recorded their discussion, summarized it with AI, and posted it straight to LinkedIn. There's so much AI content and information that when you save something and open it later, you often think, "What is this?" One example: they clicked on a short claiming someone ran 550 agents to build something huge, and there was no actual product. The content itself was made with AI, so it's AI fighting AI. They heard TikTok hasn't adopted much AI yet. On LinkedIn something is declared dead or over every day, five things a day, which is extreme. AI-written posts are getting bad enough that people can spot them. For collecting information, they read Threads and X.

The group led by someone working on AI video, character IP, virtual humans, and immersive media art included a fashion designer who makes laptop bags, people who work AI-native in the cloud, and a designer. Their conversation came together around the idea that we're moving from the prompt era to the workflow era. Being good at AI used to mean being good at prompts. Now there's a big gap even between people using the same model, and what decides the result is data structure, production process, and workflow design more than the prompt. In AI video, character sheets, style guides, references, and camera design determine quality, so future competitiveness depends more on what pipeline you design than which model you use. Many of them work while comparing Claude Fable, Opus, GPT, and Codex. Writing ability has leveled up across models, but in system design, service planning, technical docs, and architecture, the differences between models are still clear. The leader ran Fable the day before. It definitely used a lot of tokens, and they used it to develop existing prompts instead of building something new, so the payoff wasn't big, but they planned to rebuild from scratch that day. They gather insight from papers, AI media, developer communities, and real projects, and also watch how people use AI at events like this.

The group led by Byungwook Kim, who manages projects in overseas development aid, was a mix: a longtime founder, someone from a genetics company, a teacher, a legal services firm, and an IT company. The first topic was the Fable model. They talked about how fast things are moving, from prompt engineering to context engineering to harness engineering. Talking about Codex and harnesses, they discussed how to stay secure when using agent tools. Someone said burning a lot of tokens is important when you're learning, and someone joked about getting hired at a company like Teddy's Playco to use tokens. They get insights from LinkedIn and YouTube. Getting information online is easy, so the important part is picking out what matters.

The group led by Juheum Yoon, who leads an overseas AI business, included someone doing AX (AI transformation) in manufacturing, someone who worked at Apple and is now doing an MBA, a non-developer who comes to events often to learn AI, a backend engineer preparing to start a solo company, and an SRE (site reliability engineer). The SRE is most interested in agents like Hermes, because AI sessions aren't saved and your information disappears when you switch tools. Coding agents perform well, but they wonder whether human understanding can keep up. The person doing AX in manufacturing said their CEO got obsessed with AX, bought 20 Mac Studios, and keeps asking why they can't do digital transformation, why they can't just use Claude. There's a gap between how management sees AX and how people doing the work see it. Juheum came to an AX event two weeks earlier for the first time and, for the first time, considered the idea that their manager could be an AI. They had thought about specific skills being replaced by AI, but never that the managing role itself could be. Managing people is the hardest part, and it's hard even for a human, so they keep wondering how AI could do it. Since that session they've been thinking about how to bring AX to their team.

The group led by Hyungu Kang, a PM at Enconode, which builds AI services for games, included people from planning and strategy, pure NLP research, someone who moved from medical into AI, and someone who does LLM chunking. They mostly discussed how they gather information. Everyone goes where the AI information is: subscribing on YouTube, getting newsletters, or meeting up with friends. With so much information out there, everyone was converging on the question of where the people with accurate information and real insight are. Hyungu's recent insight was harness engineering. A harness is a dog's leash, and AI is like an animal that doesn't come the way you want. The key is getting it to come where you want, and reading a post about that helped Hyungu understand why data structures are set up the way they are. Their uses of AI varied: some verify incoming information with AI, some build up data in Obsidian and pull it when needed, and one person found Obsidian wasn't enough and went as far as chunking and structuring data on-premises to build their own expert. On the AX side, they talked about how to spread AI to the people around you when you love it. Ideas ranged from quietly slipping it into artists' workflows to converting people one at a time, but there's a lot of emotion involved and they found it hard.

The group led by a college sophomore included another college student, an app developer, and a developer. Someone shared their recent experience with Claude Design. With two college students in the group, they talked a lot about AI in HR and enterprise use: that junior roles are thinning out, and that, in response, junior hiring has recently ramped back up in a big way. They get information through social media, mostly LinkedIn, plus Threads and X. What stood out was one person's algorithm: scrape all their X posts, feed them into Claude, refine them through back-and-forth, then analyze the themes they care about and go find more information.

Another group had three engineers, a data researcher, an AX assistant, a manager, and a solutions architect. When the data researcher asked Fable, released the day before, about their own research area, the guardrails were so strict it got blocked. Someone on X said that if you're a researcher and a question like that gets blocked, you might be working on a topic AI doesn't cover yet. It was annoying, but they figured it meant they were studying something good. They collect information from X, LinkedIn, and curated news posted every morning in KakaoTalk open chat rooms. One person said content changes so much from three months ago that a lot of it goes stale, so they've started just looking for whatever is fun and interesting to them. When Teddy said writing matters, everyone felt a little guilty, since they're having AI do even their writing. They talked about how to write themselves, and ended up concluding that work comes down to time. You can bring craftsmanship to side projects, but at work responding fast matters, so using AI there makes sense.

The group led by someone from Tuida, a conversation service for foreigners, included two people from that team, someone running a different business, someone in business development at Salesforce, someone who travels the world a lot, and a developer. A friend of theirs showed up, so they didn't get to the second question. Their insight: people always focus on technology, but watching how Google I/O, WWDC, and Salesforce are responding to the AI era, what matters for people running businesses is how to combine AI with existing products and turn it into a business. You have to think about combining the technology with what you already have to generate real product revenue, more than about the technology itself.

The group led by Hajung Yoo, who works in sales at AWS, included a founder, three developers, a VC, and someone in sales. A fun insight they shared: one developer built a drawing skill for non-developers at their company to use, and they talked about Ansona, a model Naver built. Both are agents that let non-developers use skills, or let different roles like developers, non-developers, and designers collaborate. In the fireside chat Teddy said one person can play many roles with AI. Flipping that around, they said, there's probably also a part AI can play in helping many people work closely as one unit, and that would be fun to explore. They talked too much to get to the second question.

A group of people who are founding a company, preparing to, or working at startups talked a lot about how to use AI. Thinking about what problems to solve with AI, they kept ending up at the conclusion that you have to go offline. Online information isn't enough, and what comes out on places like X may even be slow. To find the problem you want to solve, you need to go into the field, ask users what their problems are, and find the problems a particular region and market have. That conclusion lined up with Teddy's message: build things yourself and immerse yourself.

Originally written in Korean on Brunch, June 15, 2026.