Agentic Payments
Agentic payments are money moved by software under scoped consent, the fastest-hardening layer of the agent stack.
What agentic payments are
Agentic payments are money moved by software on a human’s behalf: the rails, credentials, and permissions that let an AI agent complete a purchase, with the buyer’s consent scoped in advance instead of clicked at checkout.
Why agentic payments matter
The whole agentic-commerce promise stalls at the payment step, an agent that can shop but not pay is a research tool, and letting software spend raises the questions cards were never designed to answer: who authorized this exact purchase, within what limits, and who eats it when the bot got it wrong. The card networks and payment providers have launched agent-credential programs to answer them, making this the fastest-hardening layer of the agent stack.
How delegated spending is being built
- Scoped mandates: the human pre-authorizes bounds, merchant, category, amount, time window
- Agent credentials: tokenized payment identities issued to the agent, not raw card numbers shared
- Verification: merchants and networks confirming the agent and the mandate behind a transaction
- Audit trails: every agent purchase traceable to the consent that permitted it
Frequently asked questions
What happens when an agent buys the wrong thing?
The unsettled question the mandate system exists to shrink: tight scopes and confirmation steps for unusual purchases prevent most errors, and traceable consent decides disputes after. Merchants should expect dispute rules to keep evolving and keep their own records of what the agent asserted at checkout.
What should stores do about agentic payments now?
Stay compatible rather than clever: modern, tokenization-friendly checkout through mainstream processors, no flows that require human-only steps like image puzzles at payment, and attention as providers roll out agent-transaction support. The store’s job is being easy to pay correctly; the rails are the networks’ job.