Transaction flow agreed with our banking partner

The core question for any agent banking product is deceptively simple: where does the money live, and how does a card get charged? This month we settled it with our banking partner. Companies fund a business account on licensed European rails, and every agent card draws from that account. No card on card funding, no acquiring complexity, no detours.

The second decision was the card model. Instead of minting a new virtual card for every transaction, each agent gets one permanent card that rests frozen at limit zero. When the policy engine approves a payment, the limit is raised to the exact amount, the payment runs, and the card freezes again. Single use safety without single use cards, and far fewer issuing events on the partner side.

What is left is the detail work that makes or breaks the experience: which card controls the API can change in near real time, how webhooks confirm settlement, and how our reconciliation matches intent to charge. That is the current conversation with the partner CTO, and it is exactly the kind of boring that good financial infrastructure is made of.