A现场 at the Four Seasons Hotel Seoul on March 9. Lee Sedol, a ninth-dan Go player who played a historic match 10 years ago against AlphaGo, an AI developed by Google DeepMind, sat in front of a Go board again. This time, he met it not as a rival but as a colleague.

Using only voice commands, he completed a Go AI in about 20 minutes and even played a match against it.

At an event titled "Declaration of the Era of AI Collaboration" hosted by AI startup Enhance, Lee demonstrated a scene in which he used Enhance AI OS to autonomously carry out the entire process from planning to deploying a Go AI through voice-based vibe coding.

Enhance AI OS is a multi-agent orchestration platform in which specialized agents collaborate by combining external AI models such as those from Anthropic, Google and Black Forest Labs.

Lee's process of using Enhance AI OS was simple. All he did was talk with Enhance CEO Seung-hyun Lee (이승현), saying things like, "Go is so difficult that beginners give up. What if there were an AI teacher?" and "At first, it would be good to teach 9x9 Go."

The AI OS then analysed the conversation in real time and extracted requirements. In a method where 5 agents took over tasks sequentially, the following proceeded in order over about 20 minutes: ontology-based personalisation, PM agent research and planning, design agent draft output, coding agent code writing and OS agent deployment.

The background that made this possible was an ontology. An ontology is a knowledge representation technology that structures specific domain knowledge and relationships to help AI understand context. Enhance had pre-structured a Lee Sedol knowledge map ahead of the event with 133 objects, 156 link types and 99 attributes. It explained that it included Lee's Go philosophy, education interests and the context of his major remarks.

The OS agent analysed the pair's conversation, extracted requirements and stored them in the ontology. The PM agent searched about 100 items across Wikipedia, GitHub, Amazon books and YouTube to complete a planning document. Lee's spoken responses were reflected as they were, including "with a modern and sophisticated feel" and "so that it can also show the score and move record."

The design agent simultaneously commissioned drafts from 3 image generation models, and Lee chose a draft made by Nanobanana 2 based on Google Gemini 3.1 Flash. The coding agent implemented the app in HTML and JavaScript using Anthropic Claude Opus as the engine. It also connected an open-source Go AI engine, adding 19x19 and 9x9 modes, move-reading recommendations and position evaluation. When the OS agent displayed the finished app on the screen, Lee was able to start a match immediately without additional revisions.

Lee described the process by saying, "I also briefly experienced vibe coding. But I hit a wall. When you don't know anything, it isn't easy. But with this, isn't it basically true that I didn't do anything?"

Computing power that has advanced over 10 years supported the smooth demonstration. In 2016, AlphaGo mobilised 1,920 CPUs and 280 GPUs. The hardware that ran the Go AI on the day was only a single NVIDIA DGX Spark machine.

◆Live match: "A picture humans cannot win ... it must have surpassed AlphaGo"

After the app was completed, Lee played a match himself. Lee took black stones and the AI took white. During the game, Lee analysed the position and said, "Even if it may not feel like the highest-performing AI out there now, personally it seems to play at a level that is hard for humans to beat. This seems like a picture humans cannot win."

He explained the AI's opening style. "The AI killed all these stones, but its territory is very good. A human would not kill something like this. If you consider development potential, black stones are extremely limited. White's development potential here is enormously big. You can think of this as the difference with AI." He was explaining a style in which AI prioritises long-term potential over short-term gain.

As the match was aimed at building a Go model for training, it ended at a demonstration level. Lee also brought up his experience of retiring after losing to AlphaGo 1 to 4 and playing against AI alone at home. He said, "The AI had 20 seconds, and I had unlimited time. We played a very fair game, but honestly there were times when it took a really long time and I could not even play a single move. I remember feeling a wall that even if I played with unlimited time in front of the Go board, it would not work, that I could not win no matter what."

Lee said, "Rather than at the level of AlphaGo at the time, it would be right to see it as having surpassed that level," adding, "My basic feeling is that it has an ability that makes it a bit difficult for humans to win."

During the game, the Go teacher function provided real-time explanations of moves, prompting Lee to marvel. The AI offered advice such as, "It is advantageous to take the corner first," and "If scattered stones connect with each other, they become safe like building a solid fortress."

◆"AI has no personality, emotion, story ... a partner for collaboration"

Enhance explained the purpose of the event by saying, "We wanted to flip the 2016 narrative that humans lost to AI into a narrative that humans make AI." Lee said of AI development over the past 10 years, "It really has been 10 years, and it seems to be changing really fast," adding, "Even 3 to 4 years ago, I could not even imagine being able to do this. I thought something new would come out after AlphaGo, but the speed and destructive power after the change came were rather bigger than I thought."

Even so, Lee coolly pointed out AI's limits. He said, "As we live, memories with the other person and emotions all go into each move. And we take responsibility for that. AI has no such thing. Personality, emotion and story do not exist. It plays efficiently."

He added, "There are parts AI cannot do, and there is a realm that is only ours as humans. I think when we collaborate with AI like now, we can really show something different."

On the Go teacher function, he said, "It is not amazing technology for AI to play Go well. Rather, isn't the teacher role even more surprising?" But he added, "There is still some way to go. If there had been more time, a much better form would have come out."

Anthropic, Nvidia and Microsoft took part in the event as official sponsors. Enhance has global companies such as Samsung Electronics, P&G and Philips as clients and has raised cumulative investment of 29.5 billion won. It was selected in 2025 as the only company in South Korea for the Palantir Startup Fellowship and ranked No. 1 in commerce in the global web agent benchmark Online-Mind2Web.

Enhance CEO Seung-hyun Lee (이승현) said, "If the AlphaGo match symbolised competition between humans and AI, AI is now becoming a collaborative partner that solves problems together with humans." He added, "We will standardise ontology, agentic AI and large action models (LAM) to open an era of AI OS in which companies around the world collaborate with agentic AI."