The Founder of Circle Is Building a Bank for AI Agents. The Timing Is Not an Accident.

Every major financial infrastructure shift in the past twenty years started with the same insight: the system wasn't designed for the new category of actor that just arrived. Sean Neville is making that bet again.

Neville co-founded Circle in 2013. Circle's USDC became the financial rail for a generation of crypto-native applications, settling over $12 trillion in 2024 alone. His pattern: when a new category of financial actor emerges, the infrastructure lags by years. He built for crypto wallets before crypto wallets were mainstream. He's building for AI agents before AI agents have bank accounts.

Catena Labs raised $30 million in a Series A led by Acrew Capital and a16z crypto, with General Catalyst, Breyer Capital, and QED participating (May 20). The product is a financial permissions layer for AI agents. Humans define the guardrails — spending limits, approved payees, balance caps, counterparty restrictions — and AI agents operate within them autonomously. The agent doesn't just read your financial data. It moves money on your behalf, within boundaries you've defined.

The bank charter application is the most structurally telling detail. Catena has filed for a national trust bank charter in New York from the OCC — the same regulator that conditionally approved Mercury's national bank charter in April. A company less than two years old applying for a charter is unusual. But it's the only structure that makes sense at scale. Agents that hold and move money need somewhere to hold it with legal standing. A partner-bank arrangement creates the exact structural ceiling that Mercury spent six years trying to break through.

The timing reflects a specific technological inflection. Multi-agent workflows are proliferating faster than the financial infrastructure designed to support them. Autonomous assistants managing travel, vendor payments, expense approvals, and payroll each need financial identity — not just API access to a human's bank, but the ability to hold balances, initiate transactions, and operate within compliance frameworks that today require a human counterparty. The existing financial system was built on the assumption that the actor is a person. Catena is building for the assumption that the actor is a workflow.

The Brazil pattern here is direct. Pix already gives 170 million people instant payment rails with near-zero latency. Open Finance gives 60 million consented users the ability to share financial data across institutions programmatically. The next layer — programmable, agent-accessible financial accounts with human-defined permission structures — isn't built yet. The founder who builds Catena's architecture on top of Pix and Open Finance creates the most capable agentic finance layer in a market that already has the payment velocity that the US is still building toward.

Whether the OCC approves Catena's charter in 2026 or 2028, the category they're naming will become real. The question is whether the regulatory infrastructure arrives in time to match the technical one — or whether the agents are already moving money before the rails are officially built for them.