Myrealtrip × AB180: Claude Bloom, Round Two

Our second Claude Bloom event, with Seongpil Nam, CEO of AB180, and Donggun Lee, CEO of Myrealtrip.

Our first event last week, co-hosted with Claude Ambassadors, drew a lot of interest and a lot of feedback. The most common complaint: in an age when you can hear all of this online, why make people sit through an hour and a half of talks? We had worked hard with the speakers to plan it, but nothing is perfect the first time. (The original idea still came through, so I have no regrets.) We chewed on that feedback and changed things right away.

Seongpil Nam, CEO of AB180, very generously offered an office that was perfect for running a community event in small groups, plus great catering. Donggun Lee, CEO of Myrealtrip, has been sharing his company's AX (AI transformation) story on social media and his blog, and he offered to lay out where things stand more clearly, in person. So we put together round two with both companies.

The response in the room was very good. With the format improved based on round one, people seemed to really value getting together offline with like-minded people and having deep conversations. We still got feedback. When a table goes past eight people, introductions alone take more than 15 minutes, which makes it hard to go deep in the time you have. Some people also suggested that, instead of only the tables we set up from pre-event questions, we sometimes try something more random, or let people pitch their own topics on the spot when inspiration strikes and form groups around them.

I'm really grateful that people give us feedback this actively. It feels like putting a product online and having people care enough to send notes. At the same time, I feel a little pressure about what new and different thing to bring to round three. I'll keep planning as we go, taking feedback, using the speaker opportunities that come in now and then, and coming up with other crazy ideas.

I also talked with Joon Oh, who ran Startup Grind, one of Korea's first-generation startup community meetups. He said the key is to keep going, consistently, no matter what. When people keep seeing each other's faces, the events feel more welcoming and a real pay-it-forward culture forms.

That got me wanting more. I'd like to hold regular English-only meetups (for foreigners, Korean Americans and other overseas Koreans, and Koreans who work in English), invite people from abroad who live in Korea, and run similar meetups in nearby Japan and Singapore. That way we can connect with local AI early adopters and bring them into our Discord to build a truly global channel for AI early adopters. To get there, I've been on weekend calls with friends in Japan and Singapore to make the plan concrete.

Here are the key insights from round two.

1. Seongpil Nam: "Inspiration comes from people"

AB180 builds what may be Korea's best marketing data solution. Seongpil Nam, its CEO, gladly sponsored the space and catering for this event, and came up on stage to say a few words.

He said he spends about 90% of his day with AI, and that he'd been feeling a lot of Blue. Then he read my Claude Blue post and connected with what the event was about.

What stuck with me was what he said about inspiration. To do anything, you need the will to do it. That will comes from inspiration, and inspiration in the end comes from exchanging with other people. He himself got serious about Claude and Codex because of an older founder he met in the US: Taeho Oh, CEO of an app called Cubble. Watching that founder automate and rework everything about the app with AI, together with the team, hit Seongpil hard. That became the drive behind more than three months of changing how he and his team work, using Claude and Codex.

No matter how good the tools are, the will to use them comes from people. It was a short welcome, but it summed up why this event exists better than anything else that day.

2. A trauma from the mobile era

Donggun Lee founded Myrealtrip, a Korean travel booking platform, in 2012. He was 27 and started it right out of college. He used the iPhone earlier than almost anyone and thought he understood mobile well. Then he made a huge mistake.

Travel is expensive by nature. A typical purchase runs a few million won (a few thousand dollars), and a family trip can go past 10 million won. At the time, he didn't think payments that big would happen on mobile. He decided mobile was suited to things like Baemin (Korea's biggest food delivery app) and Coupang (Korea's Amazon), where people spend 10,000 or 50,000 won. He was convinced travel was something you planned on a PC with a spreadsheet open, laying out a four-day trip. Based on that call, he invested heavily in PC for more than four years.

The problem was the latecomers. Once word got out that Myrealtrip was growing, competitors showed up. Their teams were so small that a mobile app was all they could build. That turned out to be their weapon. A latecomer running on a mobile app alone caught up with Myrealtrip in just 10 months. Ten months to catch a company that had been at it for four years.

He described his regret this way: his imagination as a CEO had been poor. His biggest mistake was judging based only on what smartphones could do in 2012. So he promised himself that when the next wave came, he would imagine it in the most disruptive, biggest way possible. Without a moment of hesitation.

3. Launching in 48 hours, then the tide went out

That next wave came at the end of 2022. Watching GPT-3.5 come out and Sam Altman present it, he was sure. This was it. The next wave he'd been waiting for since mobile. From that moment, he said, what started wasn't Blue or Bloom. It was enormous pressure and anxiety.

The funny part is that every AI demo had a travel scenario. Seeing "plan me four days in San Francisco" produce an itinerary on the spot, he felt they were practically showing him how to use it in travel. Within 48 hours of watching the presentation, the team got together over a weekend and launched a live service. People inside raised concerns: performance wasn't good, security hadn't been reviewed. But still haunted by his mobile trauma, he pushed through: "I don't care, just ship it."

The next day, the 9 o'clock news called. The day after that, Microsoft headquarters reached out. It felt like he was finally making up for the mistake from years earlier.

But after three or four days, all the traffic disappeared. Travel is a high-involvement decision, and back when hallucinations were bad, it was a stretch to plan an entire four-day trip just on GPT's word. People tried it a few times for fun and then drained away like the tide going out. A few days later, other companies had built services at about the same level.

His reflection was clear. Back then he felt enormous pressure to "build something with AI," but that's not what consumers wanted. They had bolted on a feature, and the way they worked hadn't changed.

4. Changing how work gets done

After that reflection, he changed direction. Instead of focusing on building things, he focused on changing the company's execution and organizational structure. The change took four years.

In 2024, they set up a training group called the AI Lab, with the mission of making every employee good at AI. At the same time, they renamed their customer service subsidiary from MRTCX to AICX and changed its mission to transforming customer service with AI, not only for Myrealtrip but for other companies too. Like AWS, a company built for internal use now actually takes outside work.

In 2025, they introduced AI Champions. They found it felt more natural for people to ask the coworker next to them than to go to the AI Lab, so they put one champion on each team to spread influence from inside the team. That same year, they merged all engineering roles. iOS, Android, backend, and frontend became one role, and design and PM disappeared too. Everyone got a single title: "product engineer."

In 2026, right here at this event, Myrealtrip declared that last year was the era of raising AI literacy, and that the company has now turned into one that works AI native.

5. A company that works through AI

Lee's definition of an AI-native organization meets three conditions. Work doesn't run without AI. A small, elite team creates bigger impact. And the company stops when Claude goes down, the way most companies stop when AWS goes down.

That last one stood out to me. He said he saw it on LinkedIn. When AWS has an outage, most services stop. But are there any services that stop when Claude has an outage? That's the real measure of being AI native, he said, and Myrealtrip actually works that way.

Their evaluation system was striking too. Myrealtrip doesn't look at token usage, login frequency, or the number of AI side projects. Until last year those metrics counted in reviews, but not anymore. Instead, they look at how people used AI to improve the key metrics that mattered before AI: contribution margin, confirmation rate, conversion rate. Once you're past the literacy stage, the real work is using AI to push up the things that always mattered.

6. Flights built by a marketer, a business travel platform built by the CEO

The part of the talk that made my eyes go wide was seven real examples. Seven examples from seven teams, all running in real products and operations.

LuckyGlide is a feature where you enter your conditions and AI searches combinations of price, dates, and layovers to recommend the cheapest flight. It started from wanting more direct traffic so they'd pay less in fees to meta search engines like Skyscanner and Naver Flights. It was built by the head of marketing, a liberal arts grad with no coding experience, from scratch.

MRT Biz is a B2B service for booking corporate travel as if you're chatting in Slack. Lee had wanted to get into corporate travel for years, but it kept getting delayed: there'd be a backend but no frontend, or one piece but not another. So he built the whole thing himself, start to finish. The CEO personally built a service where you book a business trip by chatting in Slack and it connects to the ERP and submits it for approval.

Lee also built Korean Foodies himself. AI curates, translates, and tags the restaurant posts piled up in the Myrealtrip community, making 2,081 reviews across 240 cities searchable. The People team built Mywork, an attendance tool. The recruiters fixed what they didn't like about their old attendance solution themselves. The travel-companion matching calendar was built by the business development manager for Mongolia, travel collections by the marketing team, and FlightPricingLab by the flights business team. People who aren't developers in the traditional sense were solving problems in their own work.

7. The person who sees the problem solves it

One principle ran through all of these examples: the person who sees a problem and the person who solves it shouldn't be separated.

The AI Lab's failure story he told was memorable. At first, the AI Lab was a group of developers who were good at AI. Other teams would ask for something and the lab would build it. In the end, it became the same setup as outsourcing. If HR asked them to fix the attendance tool, the AI Lab first had to learn what attendance management even was. Even if they barely understood it and built something, when a policy changed, HR had to bring it back to the AI Lab to fix again. That couldn't work, so HR threw away what had been built and went back to the old way.

His conclusion from watching that was clear. The person who sees the problem solves it. So why can't that person solve it? They don't know development. They don't know AI. Tokens are expensive. Then the organization's job is to clear those barriers. That's how the AI Lab went from "the team that builds it for you" to "a training team that helps you build it yourself."

8. The CEO is the bottleneck

The questions in the Q&A showed how deep the room went. Asked what the blocker (bottleneck) in their AX transformation was, Lee answered with surprising honesty. Very often, the CEO is the blocker. Flip it around and, if the CEO works the hardest and leads by example, not much is hard. Ironically, he said, what a lot of employees actually struggle with is how to get the decision-makers above them to move.

There was a question about building infrastructure. When backend engineers started doing frontend, complaints came up: "It's the same as before. In the end I can't trust my own code, so I review it from scratch and checking takes even longer." Lee changed the platform team's mission. Instead of stopping bad code from going out, they focused on being able to recover in one second when bad code does go out. They moved investment away from blocking things in advance and toward cleaning up fast afterward.

Asked whether they could have shortened the literacy period, he said it would have been faster if the AI Lab had been a training group from the start instead of a team that builds things for others. In the end, it comes down to how well the final decision-maker understands why this investment matters.

9. Sixteen tables, sixteen realities

After the keynote and Q&A, we ran two rounds of roundtables. Round one was by job function. In round two, we regrouped people by their concerns, based on the pre-event questionnaire. At the end, one leader from each table came up and shared insights for a minute. That part showed the keywords that ran through the whole event.

The word that came up most was "harness." A physics researcher at KAIST (Korea's top science and engineering university) asked a philosophical question: "Are our thoughts themselves getting harnessed by AI harnesses?" Between thinking and acting there's some base level, and how much of that level we can hold onto is the same starting point whether you call it prompt engineering, context engineering, or harness engineering.

The talk from someone at LG H&H (LG's household goods and cosmetics company) got the biggest reaction in the room. AI has made output explode, but whether that leads to business results or changes in customer behavior (outcome) is a separate question. QA load has gone up, and they said they feel like they've become a "red pen teacher." So they summed up the core skill as defining what to build and why, and aligning output so it actually leads to outcomes, more than the ability to make more stuff. That lined up exactly with Lee's keynote point about focusing on impact.

At a table about company-wide AX, someone from Salesforce laid out four points. Is it enough to hand people tools, or do you need to build a culture? Does the AX task force itself become a bottleneck, since in the end you have to understand the other team's context? Shouldn't you focus on the parts that actually lead to revenue instead of automating every small thing? And what do you do with the resources freed up by automation: break down the walls between roles and keep only builders? Lee's "the person who sees the problem solves it" could be one answer, but the discussion showed how hard that is in reality.

At a table on redefining what "senior" means, a developer from Daangn (Korea's biggest neighborhood marketplace app) reportedly shared some insights. Lately, roles aren't cleanly separated, and development, planning, PM, analytics, and strategy are run as one integrated team. Questions came up: no matter how good design tools get, how far can a developer take design quality? If the marketing team plans and builds a service, how close can it get to final-product quality? The conclusion was that the era of integration will keep coming, and everyone's homework inside it is the ability to define problems and come up with creative solutions.

A frontend developer from NOL Universe (the company behind the Korean travel and leisure app Yanolja) pointed out the reality at big companies. Small organizations like Myrealtrip can have a CEO who drives hard, but at large organizations like Hyundai AutoEver or NOL Universe, the question is how you drive at all. Still, they felt the essential part of an engineer's work doesn't change much in the AI era. The main job is solving customers' problems. Coding itself isn't the main job.

At the product designer table, a sales leader's example got people talking. They trained an agent on their own sales experience, so if they say, "I only have 30 minutes, rank the clients I should target today," it produces a ranking and even automates the call script. The table agreed that even harness engineering has to have your own hands-on skill built into it.

Someone from Rebellions (a Korean AI chip startup) shared a conversation about making money in the AI era. The barrier to building a product has dropped, but turning it into revenue afterward requires raising the value of your own expertise. One line stuck: "Tokens are the cheapest they'll ever be right now." Meaning now is the cheapest time to build AI literacy and raise your own value.

Someone from a blockchain startup summed up their table: "Nobody knows what to do in the AI era, senior or junior. Those of us who came all the way here on a Saturday morning are the top 10%." A PM from AB180 said, "What matters is a manager who can balance the team's energy between Blue and Bloom." A table on security and governance shared practical concerns about evaluating AI in air-gapped networks and how far responsibility goes.

If the stories from the 16 tables had one thing in common, it was that they all came back to a question about the substance of the work more than tool performance: "So what is it we actually need to do?" People with different jobs, different company sizes, and different shades of Blue were standing in front of the same question.

10. The paradox of IRL mattering more

I'll also note a story I shared in my opening. The mayor of San Francisco recently visited Seoul briefly for a roundtable. Asked how Korean companies should approach entering the US, he said: don't go to big conferences. By the time content gets onto a conference stage, it's been refined and refined until it's already behind and has lost its punch, and you can't have deep conversations while networking there. You'd get far more insight walking into a San Francisco coffee shop and talking to whoever's there. Coming from someone who hosts a huge number of conferences, it was ironic, and it felt sincere.

I heard something similar from my American friend Teddy, who splits his time between Japan, Korea, and China. He went back to San Francisco after a long time, went to a tech conference, and said it was all useless. The insights you hear there are on the internet, and networking where you talk for 30 seconds and swap LinkedIns means nothing. Instead, sitting in a coffee shop, starting with "What do you do?" and ending up at that person's house, eating together and cooking in their kitchen, the conversations were much denser.

Now that AI digests everything online so easily, the value of what isn't online is going up. Sharing what's in our heads, refining it, and developing it together only happens offline. I think that's why 100 people came out on a Saturday morning: they wanted that kind of dense connection.

Originally written in Korean on Brunch, April 25, 2026.