Building Infrastructure for the Agentic Economy

TL;DR

• Stablecoins stopped being a crypto story and became a legal one. With the GENIUS Act passed in the US, stablecoins are real money. The law runs an 18-month rulemaking period that ends at the end of this year; from next year GENIUS is live, and companies and people can transact in stablecoins the way they do with bank money.
• Most agent discussion right now is inward-facing. USDC's distinguishing property is that it lives on the public internet — its typical user today is one of the roughly six billion people worldwide who have a phone but no dollar banking rails. Chandhok's bet: agents will turn outward too, becoming endpoints on the internet the way websites are.
• Today's payment rails break at agent scale, in three specific places: finality (a card payment clears through the issuing bank — that's probabilistic, and probabilistic finality compounds risk across millions of agents), privacy (a fully public ledger is great for auditability but leaks your balances, which makes real economic activity impractical), and governance (every assumption presumes a human is one step away and present).
• The fix is open architecture: agents running on open systems, interacting with other agents on those systems, paying in any available stablecoin — open by design and open by necessity, because that's how the web itself evolved from a handful of NNTP and FTP servers.

Key Points

Setting the stage: stablecoins are now legally money (~02:09–02:11)

Chandhok opened by explaining why a stablecoin issuer belongs on an agentic AI stage. Circle issues stablecoins — "a new type of money that exists in the world," rooted in crypto as traditionally understood (Bitcoin, Ethereum) but now real in a legal sense. The US passed the GENIUS Act, which made it so anybody can transact with stablecoins just like they transact with money. The law goes through 18 months of rulemaking that ends at the end of this year; by next year GENIUS is live and stablecoins are money that real companies and real people can use like bank money. USDC came out of that on-chain ecosystem, and he expects it to have legal-tender standing in the US by next year (wording to verify, below).

What makes USDC different: it lives on the public internet (~02:11–02:12)

He observed that the day's discussion of agent governance and agent security had been inward-facing. USDC is the opposite — open money on the public internet:

• Of the 330 million people in the US, most can get a dollar bank account and dollar payment methods easily.
• Of the roughly 8 billion people worldwide, about 6 billion have a phone but no real way to transact in dollars. Making that possible was USDC's first job.
• Those dollars don't live inside a bank. They live on chain, and the transactions are visible.

His central bet: money has already turned outward, and agents will turn outward too — becoming endpoints on the internet much like websites are endpoints, with a race to build them once the tech is right.

His analogy is FSD: self-driving was designed on the assumption that the roads won't change. What if you could redo the entire transportation system instead? That's the frame Circle applies to agents and USDC.

Three assumptions that break (~02:12–02:16)

(1) Agents don't tap to pay. You can hand your agent a credit card and tell it to spend. But imagine your agent spawns a hundred sub-agents: each one needs value, each needs governance, and each of their actions needs to be programmable. Existing rails can't do that. His deeper point: much of real-world commerce — and even sophisticated systems like Android — is built around what payments can and cannot enable. Stablecoins give you programmable money you can attach logic to, without the agent needing a bank account or a credit card. So agents spawning agents spawning agents can each have an on-chain wallet with governing rules, with all the tracking done out in the open.

(2) Finality is probabilistic. Finality means: I have paid you, and that transaction is final. When an agent swipes a credit card, the merchant only gets the money after the issuing bank clears the funds — that is not full finality. On chain, you want absolute finality when money moves from wallet one to wallet two.

Probabilistic finality essentially compounds the risk in the system. If you're imagining a future with millions and millions of agents — or more agents than humans, which is already true of web pages and websites today — you cannot have probabilistic finality.

(3) A fully public ledger blocks real economic activity. Most blockchain payments happen on a public ledger, which is great for auditability but bad for real economic value: if your wallet balance is fully visible on chain, you leak too much information to operate.

Underneath all three sits a shared assumption — that trust can live at the application layer, which becomes extremely difficult at agent scale. Every assumption in the current payment system, the system that moves all the value, presumes the human is one step away, always ready, present for governance.

The fix: open architecture (~02:16–02:17)

Circle's answer:

• Agents run on open systems and can interact with other agents on those systems.
• Payments for those interactions happen not just in Circle's stablecoin but in any stablecoin available in the market.
• Open by design and open by necessity — because that's how the web evolved, from a few NNTP and FTP servers in the early 90s to an innumerable number of websites and pages.

The product is at agents.circle.com: there's a CLI, and it's a marketplace of task-based activity where a payment as low as a millionth of a cent can get a task done. People are trying everything from "go infer this for me" to "go do this research" to "go write this piece of code."

All of this is designed to be operated by agents, for agents, and of agents.

Quotes

"Probabilistic finality essentially compounds the risk in the system." (~02:15)

What is tolerable at human scale becomes systemic risk at agent scale.

"Every assumption in the current payment system... assumes that the human is one step away, is always ready, is present for governance." (~02:15)

The axis of the whole talk: an agentic economy doesn't get to borrow the human payment rails — it needs rails that don't presume a human is standing by.