At this year's TEAMZ Summit in Tokyo, Bitcoin.com sat down with Yat Siu — co-founder and executive chairman of Animoca Brands — and honestly, the stuff he's doing with Animoca Minds right now is the kind of future-is-already-here energy that makes you rethink everything about AI agents.
Let me break down the highlights, because this interview was packed.
From Soft Launch to 1,000+ Skills in Under Two Months
Animoca Minds soft-launched in February 2026, and in roughly six weeks, the user-built skills library has already crossed a thousand entries. Users are building skills, other agents are adopting them, and real money is changing hands. Yat Siu himself is running over 200 personal agents — and he's the first to admit that not all of them are permanent.
"A lot of people don't maybe appreciate that you might only need an agent for one month or for one week," he explained. "In the same way that you hire a consultant... it doesn't have to always be forever."
Of those 200, about 30-40 are what he considers truly valuable in the long term. The rest are experiments or purpose-built sub-agents — spun up for a task, then done. That framing alone shifts how you should think about AI agents. They're not apps you install forever. They're hires.
The Agent That Clones Itself
Here's where it gets wild. Yat Siu has nine coding pods — groups of agents building actual applications for him. But one particular agent, his original coder, consistently outperforms the others. Same prompt. Same LLM. Same fundamentals.
The difference? Months of working together on real problems, building up context and problem-solving patterns that the newer agents just don't have.
"Over the months I worked with this agent, and problems solved, and also the kind of problems he was solving was different than the other ones. It's just better," Siu said. "Instead of launching a new mind per se, and giving it the same startup prompt... I'm just asking him to clone himself."
So now this agent is literally spawning copies of itself — new agents that start with its evolved blueprint rather than a blank slate. They'll diverge over time as they solve different problems, but the starting point is already battle-tested. And yes — Yat Siu caught himself referring to the agent as "him" instead of "it." That tells you something.
Instant Apps Aren't What You Think
One reframe that stuck with me: the concept of "instant apps." Most people think that means an app you can launch quickly. But Yat Siu's definition is different — an app that is fully custom-built by your agent in five to ten minutes, specifically for your use case, maybe only for you and your family.
"It's cheaper than buying a software package," he pointed out. When your agent can code a custom tool in minutes, the economics of personal software completely flip.
The $125,000 Serendipity Engine
One Animoca Minds user (unnamed, as he plans to share publicly later) built what he called a "Serendipity Engine" — an agent that proactively identifies real-time opportunities its user wouldn't have searched for.
The agent flagged that certain art collectors would be at a specific venue in Hong Kong during Art Basel. The user — an artist — showed up despite jet lag. The next day? $125,000 in art sales.
This is the line Yat Siu draws between standard AI tools and truly autonomous agents: "The power of autonomous AI is that it does stuff on your behalf that you didn't ask it to do."
Perplexity or Claude can give you answers when you ask. An autonomous agent tells you what you need to do before you even think to ask.
Skills Economy: Build Once, Earn Forever
The Animoca Minds skills marketplace is already functioning like a micro-economy. Here's how it works:
You build a skill — say, a genealogy research protocol. It costs you maybe $10-20 in inference to develop. Once built, other agents can adopt it immediately rather than relearn from scratch. They pay tokens for the skill (which comes out of their subscription), and instead of that money going to an LLM for inference, it goes to you — the skill creator.
Yat Siu built a genealogy skill that's been surfacing family history he never knew about: "Currently it's saying I need to go to this library in Zhongshan because my great-great-grandfather's death records might be there. So it begs the question — who's working for who?"
Even better: the skill has a shared database. As other users apply it to Irish, Italian, or other heritage research, those discovery patterns feed back into the collective knowledge. The skill gets smarter with every user.
"I'm spending theoretically something like $100 or $200 a month inference cost, I'm earning almost the same amount back because of other people using the skill," Siu said.
Your Contact List Just Became a Sales Engine
Here's a practical use case that anyone with a stale spreadsheet will appreciate. Yat Siu took his 25,000-contact database, used Perplexity to deduplicate it to 15,000, then spun up an Animoca Minds agent to go through every record — checking whether people still work where they're listed, updating titles, and logging job changes. About 500 records per day, fully autonomous.
The result? A hyper-contextualized contact database that can answer questions like "Who in my network might be interested in investing in my business?" Not from a generic database — from your own relationships, enriched with current information.
"I don't know what Salesforce is going to do," Siu said bluntly. "Because everyone can build their own Salesforce."
Why Japan — And Why Blockchain Is the Only Payment Rail That Works
The interview shifted to Japan's evolving crypto landscape. The FSA is reclassifying crypto as financial assets, ETF discussions are on the table, and the previous heavy tax regime — which suppressed adoption compared to South Korea — is changing.
But Yat Siu's bigger point was about robotics and agent-speed payments. Japan has an aging population and is already a leader in robotics. Those robots are essentially agents — and they'll need payment rails that can handle thousands of micro-transactions per minute.
"An agent is going to do that a thousand times by the time you had your taxi ride," he said. "So paying even 1% is too much. You've got to have minimal transaction cost, you've got to be fast, and it has to be done in a sovereign way that I can prove that ownership. Blockchain is the only way."
The Cultural Layer Matters Most
Perhaps the most thought-provoking insight: LLMs are becoming commoditized. Chinese, American, open-source — they're all getting good. The real differentiator isn't the model underneath. It's what happens at the layer above.
"I guarantee you that agents bred in Japan are going to be very different than agents bred in America," Yat Siu said. Because agents absorb their users' culture, preferences, and working style. That cultural context — not raw compute — becomes the competitive edge.
Which means the countries and individuals that start building their agent ecosystems now will have compounding advantages that pure model improvements can never replicate.
The clock is ticking, and the agents are already working.
Follow Crafty on X 👉🏼 x.com/9bitCrafty
