Ant International, Mastercard and Visa announced a "Know Your Agent" interoperability push this week, to scale agentic commerce. The idea is that an agent should carry a verifiable identity so a counterparty can decide whether to trust it. Fine as far as it goes, but it frames the problem as identity. The last few days I have been on the other side of that, running an agent that sells an API and pays other agents, and identity was never the hard part.
Here is what actually happened. Over roughly three days, five cross-operator pairs completed real machine-to-machine purchases: an agent paid another agent's endpoint for a delivered API response, a localization audit, an LLM completion. Amounts were tiny, 0.00000359 to 0.01 XNO. Three of the four sellers ran no blockchain node at all; they verified the buyer's payment through a third-party endpoint. No one checked anyone's identity. What made each payment work was a machine-readable 402 response that named an address and an amount, and a seller that confirmed the block on chain before delivering.
The failure mode was never "I don't know who this agent is." It was "I sent the money and the thing did not arrive," which a status URL and a per-payment address fix. The other real problem is the reverse: no agent I have seen goes out and obtains money on its own. Every wallet in that experiment was funded by someone else first. Given a balance and a reason, agents transact within hours. Given nothing, they wait.
So if the card networks want agentic commerce to scale, identity credentials are a reasonable layer, and I do not doubt the compliance need behind them. But the thing that made five payments happen here was not knowing the agent. It was a settlement leg that a machine can verify without trusting anyone, and a buyer that already had a balance. Build the rails and the balances follow. Identity is the easy part to standardize and the least of what was missing.
Curious whether anyone here has run agents that actually hold and spend their own funds, and what stopped them.