Probably one of the most fascinating panels tonight at @MetaMask Builder Nights Tokyo. The conversation brought together perspectives from the @ethereumfndn, @MetaMask , and other builders working across wallets, identity, AI andonchain infrastructure - essentially asking what the financial experience looks like when the next user may not always be a human, but an agent acting on a human’s behalf. The panel started with a simple question: How many people in the room would trust an AI agent to execute a financial transaction on their behalf today? From where I was sitting, maybe ~10% of the room raised their hands. That was interesting because the technology is already moving ahead of the trust curve. One example discussed was a firm in New York already using AI agents across transactions involving traditional assets and Bitcoin, while keeping a human in the loop. So adoption has "started"... but we're still very much in the baby steps, and resistance is understandably strong. Another part of the conversation that stayed with me was stablecoins. We've spent years talking about how blockchain and stablecoins can remove intermediaries and make money move faster. But there's another side to that equation: the faster money can move, the faster mistakes can move too. Whether the mistake originates from a human or an autonomous system, programmability and instant settlement can compress the time available to detect something wrong, intervene, and reverse course. Every technological shift comes with that trade-off. We gain efficiency and remove friction, but some of that friction historically functioned as a control layer. This is why I particularly enjoyed hearing MetaMask's perspective around wallets and identity. As we move further into the emerging #AiFi era, I don't think the question is simply “Can an agent transact?” The harder questions become: Who is the agent? Who authorized it? What is it allowed to do? What information is it making that decision from? And who is accountable when something goes wrong? From a policy perspective, I think this is where things become especially interesting. Builders naturally optimize toward capability: make agents faster, more autonomous, interoperable and useful. Regulators and financial institutions necessarily approach the same system from another direction: authorization, accountability, auditability, consumer protection, financial crime and systemic risk. Those worlds may diverge considerably before the infrastructure eventually forces them to converge. And somewhere between them, we're going to need a new trust stack. This is also something I've been thinking deeply about as @KnightIntelco work with @tryforgeintel on KYA — Know Your Agent. My hypothesis is that we're going to see an entirely new category of infrastructure providers emerge around agent identity, authorization, security, financial intelligence and trust. There may eventually be a handful of "dominant providers" in some of these layers, much like we've seen large model providers such as @OpenAI and @AnthropicAI emerge at the foundation-model layer. But identity alone doesn't solve the entire problem. Agents also need context. An agent can only reason and act from the information available to it. Input → reasoning → output → execution. If the financial context feeding that system is fragmented, stale, incomplete or poorly understood, giving the agent permission to transact doesn't magically make the underlying decision intelligent. In finance, that's particularly consequential. Financial identity and behavior are already fragmented across banks, wallets, jurisdictions, payment rails and onchain activity. An agent operating from incomplete context can produce a perfectly executable action based on an imperfect picture of reality. And once execution becomes autonomous and money moves at machine speed, the cost of bad context gets much more interesting. So perhaps the next chapter isn't simply AI agents + payments. It's building the identity, intelligence, authorization and accountability infrastructure that makes autonomous finance trustworthy enough for people and institutions to actually use. Still early. Still lots of unanswered questions. What a fascinating time to build. 🇯🇵⚔️