Myrealtrip × Jocoding: Claude Bloom, Round Three

Three weeks ago we co-hosted an official Claude community event with Anthropic's Claude Ambassadors, and since then I've somehow been turning into a community guy. People keep coming and telling me they had a meaningful time, which always gives me energy. At the same time, I keep wondering what value our Claude Bloom community can give the people who come, and what benefit it can create for my company. It has to be a win-win for both sides.

So far I've had two ideas. The first is a stepping stone for overseas go-to-market (GTM). The second is a platform for "bidding" on AX (AI transformation) projects.

It would be great to use local communities to break into overseas markets, but they ask for a lot of sponsorship money and still don't let you promote freely. Part of me thinks I should just build a community in that country myself, but I can barely do it in Korea, so how would I do it abroad? Then Claude Bloom took off by chance and a community formed here in Korea. So I thought: why not package the good examples from here and launch them overseas? To act on that right away, I explained what we'd done in Korea and the idea of Claude Blue to the Anthropic Claude Ambassadors in Singapore. They said, "Claude Blue, that's something all of us feel," and approved a collaboration on the spot. So our first overseas Claude Bloom event will probably happen at the end of May.

Second, since Cliwant (my company) is an AI bidding startup, we get a lot of enterprise AX requests. Sometimes, when we look at a proposal, it's a project that's awkward for us. Taking it on would mean a lot of R&D and probably no ROI, but turning it down feels like a waste. After that kind of half-useful, half-useless situation kept repeating, I thought again: if our strength is bidding, what if we took the tough processes and methods of public-sector procurement and used them to match large companies with AI companies in the private sector? Large companies would award projects transparently, through competition, at the lowest bid. AI companies would win high-ROI projects with minimal upfront sales and networking. And we could charge a fee in between.

The Claude Bloom community actually has a pretty high share of enterprise people. By our numbers it's about 18%. The bigger the company, the more FOMO about AI, and other AI communities lean heavily toward developers. I think that's why people gathered around a topic like ours, which is accessible to everyone and also unique. So the community is very broad.

Honestly, there isn't much we can give the people who come. There's plenty of insight online, and coming here mostly gets you networking, which isn't enough on its own. So I think it would mean a lot if AI companies and users from many countries, the US, Japan, Singapore and others, not only Korea, joined our Discord channel. It would help people get a read on how AI is landing overseas, do GTM, or find partners. At least for us, I think it would.

Large companies also don't really know which companies are good at AI, so they have to verify through PoCs. If they could compare and check that right away, I think that would be a real advantage. For AI companies, it goes without saying. I brought this up at the event today and the response was good.

Finally, I thought about how to describe Claude Bloom's identity and settled for now on "The most Vibrant and Honest AI community." Requests are coming in from all over, so we hold a community event once a week, talk often on Discord, and small groups are forming, so "vibrant" seems right. "Honest" is because our whole theme starts from the Blue caused by AI and is about growing together into Bloom. So it becomes a place where people share the problems they're running into and look for ways through together, more than a place to show off. Everyone is meeting for the first time anyway, so there's no need to put on a front.

I don't know how long this community will last, but I'm going to take it as far as it goes. Beyond events with speakers, I'm thinking about more fun formats too, like an outdoor AI hackathon, an AI morning coffee rave, and a Han River walk meetup (walk instead of work).

1. Myrealtrip's offer

At the last event, round two, after CEO Donggun Lee finished his talk, Myrealtrip reached out first. They offered to lend us their space and cover catering. That's how our third event came to be held at the Myrealtrip office, less than three weeks after Claude Bloom started.

For the first event, 2,000 people applied. In three days. We couldn't host them all, so we invited only 100, and the disappointment of the other 1,900 kept the events coming one after another. For this one, our speaker was Jocoding, an AI coding creator with 730,000 YouTube subscribers.

2. The most vibrant and honest AI community

In my opening, I described Claude Bloom's vision this way: the most vibrant and honest AI community. Vibrant means bringing good people together in good spaces and making fun events. Going forward, we plan to try unusual formats too, like a morning coffee rave or a hackathon in a park.

Honest is a bit different. Claude Bloom started from AI Blue: the fear, the sense of loss, the feeling of falling behind because of AI. Since the people who show up relate to that, it turns into a place where people bring out the problems they're facing and look for ways through together, and nobody comes to show off. Everyone is meeting for the first time, so there's not much to hide.

3. What Myrealtrip changed after 15 years

Wonjin Heo, Myrealtrip's CTO, gave a talk. Myrealtrip is an independent travel platform with more than 10 million cumulative sign-ups and about 260 employees.

Heo talked about customer service. They get 75,000 inquiries a month, split about evenly between phone and chat. For 15 years, that volume grew as the platform grew, and each time they handled it by hiring more people. Labor costs naturally went up, and agents didn't stay long. Frequent turnover piled up training costs and dragged down the customer experience. Heo said they put up with that vicious cycle for 15 years.

Then, early last year, they started bringing in AI.

4. Same setup as before, with a chatbot on top

Myrealtrip was careful about how they brought it in. They didn't replace everything from day one. They kept the chat solution they were already using and added a chatbot only for simple questions that could be answered from the FAQ and policies. Agents didn't have to learn a new system, and anything the chatbot couldn't handle naturally went to a person.

When they saw results, they connected APIs, and after that they expanded step by step to AI agents that take action directly. They started with simple things like resending password reset emails, and AI began handling complex cases too, like calculating refund fees.

Early on, the rate of compensation paid out for AI mistakes was about the same as for human agents. Now, Heo said, the AI's error rate is much lower than people's. As a result, the number of agents handling simple inquiries went from 57 to 28, about a 50% drop. The people who moved off weren't let go. They moved to higher-level support, handling emotional complaints and urgent cases. They also confirmed that the quality of AI support isn't much different from human agents.

5. From 25 a week to 5,000

After CS automation, Heo moved on to operations automation. When they brought in products from overseas, translating them, matching categories, and organizing them by city was all done by hand. Twenty-five a week was the limit.

Now you press one button and enter a city, and AI analyzes trends, picks categories, and automatically lists products at 4 a.m. It can handle up to 5,000 a day. Cities they never used to touch, in places like Africa and the Middle East, started getting products listed, and real sales came in from Cairo and Monaco.

6. Jocoding and the animal face test

Jocoding's talk was "Solo founding in the AI era."

In 2020, he built an animal face test. You upload a photo and AI tells you whether you have a cat face or a dog face. That's the whole feature. It hit No. 1 on Naver's real-time search rankings (Naver is Korea's biggest search engine), beating out celebrities. It was shared by 120,000 people in Instagram Stories and used by more than 40 million people in English-speaking countries. It made the foreign press. It got as high as No. 2 overall on the App Store in Brunei.

Back then, ads were the only revenue. Some days AdSense hit 6 million won, but he never added paid features. Looking back, he said it's a huge regret: no paid option at the peak of the viral moment. Better late than never, he's now added Stripe payments and a full report feature. Since its 2020 launch, without any changes, the service still brings in 650,000 to 1.2 million won a month.

7. Zero operating costs

The other point about this service was operating costs. It's a static website on Cloudflare Pages. Bandwidth is unlimited, so even if everyone in the world visited at once, it wouldn't cost anything. The AI is on-device, built on TensorFlow.js, so there's no server. It runs right in the browser. Other than the domain, there are no operating costs.

His subscribers built things the same way. Services like a face-reading test and a first-impression test came out, and one hit No. 1 on real-time search three days in a row. One subscriber made 100 million won with a personal color analysis. A coding beginner made 10 million won in three days.

8. One person doing everything

Sam Altman's comment that we'll see a one-person unicorn made headlines. Jocoding said he felt it could really happen. His evidence was the vending benchmark. When a human and an AI were each given a vending machine to run, the human made $844 and Gemini made $4,387. His point was that AI has started to beat humans at business judgment too, not only coding.

We've reached the point where one person can do planning, development, and marketing. Technically, a large company and a solo founder use the same AI models. Whether it's Claude or OpenAI, everyone is wrapping the same models and using them. The tech gap has basically disappeared. What's left is the ability to design. Overseas, there's already been a case of a one-person company selling for $80 million, about 120 billion won.

9. The product builder factory

So Jocoding built a product builder factory: a setup that gathers people who can each do planning, development, and marketing alone, and keeps cranking out products. When he posted the job at the end of last year, 280 people applied and he hired three. The terms were a guaranteed base salary of 40 million won plus 10% of revenue as an incentive.

Right now those three, plus a junior he already worked with, four in all, are churning out a range of products. Asked about the results, he answered honestly: "Nothing has really hit yet. All together it's not even a thousand dollars." It isn't easy to admit failure at an event about success, but he just said it. Since nobody knows which parameter will work, he said, they're experimenting in volume to find it.

10. When to quit your job

The most memorable question came in the Q&A. The person asking was a CTO at a mobility startup who said their personal income from development had just passed their salary. They wanted his take on whether to quit.

Jocoding said he also ran his YouTube channel alongside his job as a developer at LG CNS (LG's IT services company) for a long time. He didn't quit even after the income got fairly big. His own rule is to keep the safety net until side income reaches about five times your salary. These days, he added, you can set AI agents running and keep both going while still at your job. Make the call to quit when you're completely sure you've hit the real tipping point.

11. Sixteen tables, one question

Then it was time for the roundtables, the part I'd called the best thing about this event in my opening. Every table got the same topic: what AI can't replace, and which organizations need to become AI native.

The stories from the 16 tables were similar and different at once. One person said they let go of 60% of their development team and output doubled. Someone said labor costs dropped sharply while what they spend on Claude and Codex is under 100 million won. Someone said that at companies on air-gapped networks, bringing in AI at all is still hard. And someone said AI has cut their workload, and they don't know what to do with the time left over.

12. What AI can't replace

The insights from each table pointed in a similar direction: responsibility, judgment, intuition, communication.

A marketer at a pharmaceutical company said AI says a lot of reasonable things but sometimes gets it wrong at the decisive moment. Catching those moments is a person's job right now. Others said leaders need to become AI native first, because if decision-makers don't understand AI, the whole organization won't change. One table concluded that the human role is solving problems that go from zero to one, and telling the story of it.

Everyone was wrestling with similar questions, and that made today's gathering mean more.

The next event is May 14. Someone who worked with Elon Musk three times a week at xAI is visiting Korea briefly. They also helped build early Gemini models at Google DeepMind. It should be a chance to hear firsthand stories from Silicon Valley.

Originally written in Korean on Brunch, May 7, 2026.