Agentic payments: the rails autonomous agents need
When software pays software, the questions change. Who approved it, what are the limits, and who is accountable?
- Agents pay more often, in smaller amounts, at any hour.
- Every payment needs clear authority and an audit trail.
- Wallets, payment protocols, trust layers and reporting are the rails to build.
An agentic payment is a payment started by software acting on someone's behalf. It sounds like a small change. In practice it rewrites most of the assumptions behind how payments work.
What changes when the payer is software
Volume. Agents can make far more transactions than people, many of them very small. A payment system that charges a fixed fee per transfer, or takes days to settle, does not fit.
Timing. Agents work continuously. Settlement that pauses on weekends and holidays becomes a real constraint.
Authority. Every payment needs a clear answer to the question "who allowed this?" For agents, that answer has to be encoded in advance as rules, limits and permissions.
Accountability. When something goes wrong, there has to be a record that shows what the agent did, why, and under whose authority.
When the payer is software
The rails that need building
Agent wallets with spending limits, allow-lists, time windows and escalation to a human for anything unusual.
Payment protocols that let an agent pay for an API call, a dataset or a unit of compute in a single step, without a sign-up form or invoice.
Trust layers that let one agent check the identity and record of another before doing business with it.
Reporting tools that turn thousands of agent transactions into something a finance team can reconcile at month end.
The rails, layer by layer
Why this is in our thesis
Agentic payments sit where two of our focus areas meet: AI and stablecoins. We expect the most valuable companies here to be infrastructure, not applications, and we are actively meeting teams building it.
We read every pitch that fits our thesis.