D4D: A Night at a Defense Hackathon

I went to D4D, a wild defense hackathon hosted by Kangkeun Choi of EGCED Tech. A few weeks earlier he'd shown me the visuals for what he was planning, and the real thing matched them. Maybe it was even better. The operations and the planning were excellent, and the staff were all great people. I joined as a session moderator, and I left feeling like I was the one who got the good energy.

When you run Bloom community events, a lot of people ask why you do it. Even after I explain, they look unconvinced. Their eyes say there must be some hidden motive, or else I'm just unusually pure. Honestly, I don't get why they ask. I just do it. It's fun, haha. I meet good people, I pick up insights, and it beats sitting at home. That's the whole reason. It seems obvious to me.

The funny part is that I still wonder why Kangkeun does this. I've asked him several times, and he always says it's to raise awareness of defense in APAC, connect builders with defense, and build bigger momentum. To be honest, that still doesn't fully land for me. Then again, this is the kind of work nobody else will ever understand unless they're the one doing it. In the end, you believe in your own vision and push it forward yourself. Today I felt, 100 percent, that calm but intense, certain look in his eyes. The man is a bulldozer. He's going to make it happen, no matter what.

Usually I only run my own events, so going to someone else's as a guest made it clearer why people do this. When an event runs well, people keep wanting to ask questions and help out. And when they meet good people there, all that gratitude and respect flows straight to the host. Some events end as just events. This one was different. It may be the most efficient marketing there is: you plant trust in everyone gathered in one room. A good offline experience, once it's properly planted, stays in people's heads for years.

So I think everyone should host a community meetup at least once. It doesn't have to be big. Pick one sharp topic you care about, open a Luma page, and announce it on social media. Surprisingly few people actually host their own, so it's still wide open. It's not a big deal. Now let me tell you about the night builders and defense collided.

A night where builders meet defense

D4D stands for Deploy for Defense. It's a 24-hour hackathon that bills itself as the Asia-Pacific defense tech builder network. AI engineers, developers, designers, founders, active-duty soldiers and veterans, people from completely different backgrounds, come together to work on real problems in defense and security. I gave a community session talk and ran two fireside panels on connecting builders and defense. After dinner, in front of everyone gathered, I raised the one question our community cared about that night: how can builders and defense connect?

Put the battlefield at the center

The night opened with a keynote from the CEO of a company that works on inter-Korean and security intelligence. He joined by webinar from the Czech Republic and gave a talk titled "Strategic Deterrence: The Ukraine Model." He started by saying few would dispute that the pace of battlefield innovation has changed faster than anything else since the Russia-Ukraine war. The technology gap with adversary states is now about three to six months, he said, so a weapon you use today becomes hopelessly outdated within half a year. In Ukraine, they test and upgrade 300 to 500 pieces of hardware and software a year, turning them into what are effectively new products every three months. On a battlefield where both sides chase each other like spear and shield, and the innovation cycle has become extremely short, new technology goes into use the moment it appears.

He laid out three mindsets defense startups need. The first is battlefield-centered thinking. The goal isn't just solving a problem. You have to approach it as solving a problem on the battlefield. That means understanding national institutions along with every component of the battlefield, for both sides: strategy, doctrine, tactics, the battlefield environment, organization, command and control, training, and logistics. For example, below the battalion level, the Chinese and Russian militaries don't use the same organizational structure we do. They operate as mission-centered task forces. A solution built without that understanding might solve a tactical problem but fail to solve the real battlefield problem, or cost far too much to adopt.

The second is designed-in scalability. From the very start, you should ask how quickly you can modularize and expand survivability. He described a modular structure that scales flexibly when new features or changes come along. Build it that way, he said, and scaling again becomes much easier.

The third is a shift in perception. We aren't at war with Russia or North Korea, but we are in constant competition with them, and that competition can turn into crisis and conflict at any moment. We need to see it that way. The ability to develop solutions that can be deployed to the battlefield immediately is itself a strategic deterrent, he said, because it makes it hard for an enemy to take decisive military action. So when you design solutions as if we're at war right now, the results come out far more practical and effective.

Solo companies, where communication cost is zero

Next up was Gubong Jung of Team Attention. He got a big laugh with a story about his recent civil defense training, which he was taking online. He left the browser open and asked Codex to keep clicking the Next button for him. Codex refused, telling him this was important security material and he really needed to watch it himself. That left him two options: use a Chinese model, or click it himself. He said he clicked it himself.

The area he watches most closely is the one-person company. He thinks solo companies may be the biggest change AI brings, because what sets them apart from every other company is that their communication cost is zero. When an organization hires someone to get more done, output doesn't double. It goes to 1.5, and often even lower. That's communication cost, and you can calculate it mathematically. When you work alone, that cost drops to zero, and with AI behind you, it can grow exponentially. A company is a machine that uses a system to compound, he said. In the past, building that system meant gathering people for development, business, and growth. Now AI has both intelligence and labor, so if you have a good enough problem and you're good at using AI, you can build the system on your own.

So what you need isn't capital, he said. Companies used to need capital to hire developers, but now Claude and Codex do the development, so that capital isn't needed. It makes more sense to build the system first without capital, and raise money at the point where putting capital in would make it grow exponentially. And when people say there's nothing left after paying for the API, he said, the reason isn't that tokens are expensive. It's that they haven't found the right problem.

His real message was this. Real problems aren't glamorous at all, and not a single one gets solved by sitting in your room typing on a computer. The things you can solve in your room only earn you likes online. Real business problems show up when you leave your room, meet customers in person, and deal with much messier stuff. So if people who are great with AI are going to stop bringing the same predictable things, communities need to work properly, so builders can meet real industries like defense, shipbuilding and marine, and manufacturing. That's why a defense hackathon like this one matters so much, he said. Later, I opened my panel by quoting him: an AI builder can get to 90 points alone, but the last 10 points come from a useful collision with domain, customers, and good problem definition.

From Claude Blue to Bloom

On my turn, I shared what I've learned running a community. I work as a marketer at an AI startup, but I never set out to build a community as a strategy. I was afraid Anthropic and OpenAI might eat up our software entirely, so I started asking all sorts of people for coffee chats. One of them was an older friend from college, a senior engineer at Meta HQ, who told me AI was making them really depressed lately. They felt that 20 years of machine learning experience was becoming useless. I wrote about that conversation on my personal blog and called it Claude Blue, borrowing from "Corona Blue." The post unexpectedly went viral in Korea and on Reddit. I was curious about the people who felt the same way, so I opened a meetup, and over 2,000 people signed up. That's how the community formed on its own. When you first meet AI you get excited, and after using it a lot you get depressed. The name Bloom means not stopping there, and blooming instead. Today we have 2,400 members and more than 5,000 total applicants.

I shared a few quick insights. I opened meetups to ease FOMO, then worried I might be feeding FOMO myself. But the real reason people come offline is that they want to connect with like-minded people, and that alone was worth their money and time. People weren't interested in being good at AI for its own sake. They cared a lot more about how to solve real business problems in their own domains. Because everyone has the illusion that vibe coding lets them code, a downside has appeared too: people skip learning the fundamentals. So things seem to be heading toward a world where only the real experts survive. Fast execution matters, but sometimes the quirky ideas only humans come up with create more value. Telling apart when speed matters and when planning matters has become even more important.

An eye for good output matters too. Before AI, you learned by getting chewed out by your seniors. When you hand work to AI, you skip that process. Even if you skip it, you have to keep a firm grip on the wheel: your understanding of the plan and the output. There was also the obvious point that even in the AI era we're still human, so stamina matters. Above all, I see a dangerous gap. Developers have felt the AI wave hard since last year, but for many non-developers it still hasn't hit home. The difference comes from structured versus unstructured work. When the wave does reach you, it hits much harder, so spreading the stories of people who caught the wave first is part of why I run Bloom.

I take AI as a mindset more than a technology: the spirit of "Maybe in the AI era we can do this too." I also shared my view that people with broad yet deep knowledge will be in demand, and my definite optimism that a founder's ambition, grit, and execution matter more than market size. Because AI makes products easy to build, the bottleneck is shifting to marketing and GTM. So I said it's better to build your sales channel before you build the product, and if you don't know what to sell, try selling yourself first. If I can't even sell myself, I can't sell any product, however good it is. I added something I heard last week when we hosted the rapper Paloalto. Artists also need to make music with AI, but when we listen to music we're listening to more than the music. We're also hearing the story before and after it, so humans should focus on storytelling. Software is the same. Build the product fast with AI, and let humans focus on the product's story and branding.

A developer unit inside the military, and burgers

More community introductions followed. Jio Baek runs an alumni community for veterans of the Air Force's systems development unit and also founded Kits Labs. He told us about a developer unit of around 200 people under Air Force headquarters at Gyeryongdae (the joint military headquarters complex in central Korea). It's an unusual unit where you spend your whole service writing code, so they hold internal conferences, hackathons, and algorithm contests, and many people who leave go on to publish papers at international AI conferences or start successful companies. Israel's Talpiot gathers top talent in the military and feeds it into startups. The community started from a simple question: Korea has a capable unit like this too, so why isn't it known outside, and why can't people brag about it after they're discharged? Now, through mentoring, pro bono work, homecomings, and hackathon support, they're building a culture where the military and the private sector move forward together.

Hyungmin Jin of Cry Cheese Burger, which gave big support to that evening, also left an impression. Why does a food and beverage brand show up at so many community events? What mattered most to him was the ecosystem. When you put on an event, covering costs, food, and getting people together are all hard, and he has been supporting those parts and connecting people with each other. He compared an ecosystem to Lego. Depending on how you assemble the pieces, you get a masterpiece or something functional, so he collects more pieces and meets the people who'll assemble them one by one, offline, and connects them. Food is something people always need to eat, after all.

90 points alone, the last 10 from collision

The first panel was with four community builders: Kangkeun Choi, CEO of EGCED Tech; Gubong Jung of Team Attention; Hyungmin Jin of Cry Cheese Burger; and Sumin Chae, a former Army supply officer who makes YouTube videos as "Lieutenant Shu." I opened with Gubong's 90 points and 10 points, and said I hoped this panel would be that useful collision.

When I asked Gubong about what solo builders can do, he answered candidly that almost nobody has already created big business value alone, and most people who claim to are lying. But the people who are good at it are quietly doing well, he said, and he told us about a younger colleague he respects. In the software industry, adopting AI well means cutting people. In factories, going through AX (AI transformation) actually means hiring more, because you need people to run machines that weren't running before. Truly creating value with AI means the market pie grows and there's more work to do, he said.

I asked Kangkeun what perspective builders heading into defense are missing, and who they need to collide with. He said that in the past, large defense systems companies talked to the military on the strength of their systems. Now builders and startups are emerging in the defense industry, and their technology has already reached a certain level. The next step is domain logic. The collision point is the spark point where people who know the domain on the ground meet people who solve those problems with technology. A hackathon like this is that spark point, he said. He had watched things like the National Space Hackathon in the US and the defense ecosystems that grew on their own in Europe, but the Asia-Pacific region lacked spaces like that. So he wanted to create a place where people with military expertise and developers mix, collide, and throw off sparks.

I asked Hyungmin how an ecosystem forms. His answer was inefficiency. He has done every inefficient thing, to the point where others wondered whether he really needed to. Once an event went over capacity and an extra burger order came in, but delivery couldn't make it in time, so he ran the burgers over himself. It happened to be raining that day, so he ran through the rain. Why do the inefficient thing? Because he believes that when you keep making the people who asked for help succeed and have a good experience, it eventually comes back to you. He even measures, qualitatively, the happy time people spend eating his food. I could relate. Seeing someone else put in that kind of effort becomes encouragement in itself, and it turns into content that makes you want to cheer them on even more.

I asked Lieutenant Shu about the biggest misconception builders without development experience have when they tackle defense problems. He said the military doesn't reject technology because it's conservative. Adoption is hard because of weak reliability and serious security issues, and the field has a huge number of problems. If you request supplies you need right away, you wait 15 days. The actual quantity you have on hand can differ from what the military's integrated information system says. So builders need to listen to a lot of problems from the field to solve development problems. There needs to be a bridge connecting military terminology and systems with leading-edge technology. And if you train only on synthetic data, you can get friendly fire or hallucinations on a real battlefield, so you need a process that validates with real data.

The point of this session was clear. Developers took away the military's domain, and people from the military took away the developer's domain. It was a place for collision, with perspectives moving in both directions.

What investors and the military look for

The second panel brought the investor and military perspectives. Yeojun Cho is a partner at The Ventures and served as an Air Force officer through the college graduate officer program. Jaehee Ahn handles Navy AX for the National AI Strategy Committee, which reports directly to the President. He graduated from the Korea Naval Academy and served 18 years, working across combat systems, informatization, and logistics programs.

When I asked whether many private-sector builders have made it into the defense field, Jaehee said, to give the conclusion first, that even if they want to come in, it's hard. The bottleneck is institutional. There's no contracting system that lets engineers go into the field, and even if they offer to come in for free, security makes it hard. Domain experts can't tell outsiders everything about how we fight and how we support operations. The US and Ukraine, which are fighting wars, use things quickly without fully verifying reliability, but our procurement system works differently, he said. War gave those countries the justification to overhaul their institutions. His candid diagnosis was that Korea is obsessed with transparency, equal opportunity for everyone, and leaving no room for complaints, so it can't change quickly.

When I asked whether defense is a money-making industry from an investor's point of view, Yeojun said there's a lot of room, but the potential is still much bigger than what has been realized. He sees the battlefield, not the factory, as the place where physical AI truly happens. Search "defense" on Y Combinator in the US and you get a huge number of results, and it's spreading to personal use too, like drones that catch mosquitoes. He said now is a good time to create a defense-only fund in Korea. As Chinese and US weapons systems quickly split apart, the US needs supply factories it can trust, and no country is better placed than Korea to take that role. But regulations aren't friendly to foreign buyers coming in to acquire companies, and exit options are limited to the likes of Hanwha Ocean or Hyundai, so the potential is big but hasn't become reality yet.

An analogy about tenants and owners came up too. It's a question of who owns the final product. SK Hynix holds one position in a value chain, so it can't help being dependent. But something like the KF-21, where we decide the final product, unlike the F-15K we brought in from the US, isn't a tenant. In the end, modern war is information warfare, and how independently you run that information is what gives you the authority to decide for yourself where a product gets used.

When I asked how defense startups get through the valley of death, Jaehee said a plan to grow new defense industry companies was announced recently, and defense funds have shored up the money side. But the valley of death really comes from not being able to reach users' needs. You win contracts by being chosen by users, and if you can't reach them, you can't win contracts. Fortunately, five AX hubs have opened, so working with domain experts there should help builders find those needs.

I asked for ideas on how builders and defense could get closer. The premise was that Korea has an advantage: nearly everyone serves in the military, so it's a country full of people who know field problems firsthand. One proposal was an ecosystem where long-serving reservists and noncommissioned officers, who know the field's problems and their solutions, support startups as mentors. That would also help with jobs for reservists. I chimed in too. If young people saw the military as a place to go find startup ideas, it could actually become fun, and in the US, veterans get a lot of extra points in procurement. Yeojun said that if the private sector and defense work together on turning experience like being a drone operator into an asset, and if we get to where overseas buyers like Lockheed Martin are eager for Korean startups, a lot more good builders will come out of it.

What stayed with me after the two sessions was the simple fact that builders, the military, investors, and communities were all in one room. You can get to 90 points alone. The last 10 points come from useful collisions between very different people, like these. I hope this night was a small beginning for that kind of collision.

Originally written in Korean on Brunch, July 5, 2026.