A $5,000 request
Could proceed under standing authority accepted by the organisation and signing provider, after all required checks.
AGENTIC AI PAYMENTS ORCHESTRATION
Agentic payments.
Built around your authority.
A controlled approach to agent-initiated payments across bank, card and stablecoin rails, subject to supported integrations and customer-authorised execution.
Discuss your use caseWHY ORCHESTRATION MATTERS
Continuous commerce.
Accountable spending.
AI agents may buy data, request computing capacity or coordinate supplier payments while finance teams are offline. Each request still needs a valid business purpose, funding and an accountable owner.
Agentic payments can use fiat currency through bank and card rails, as well as stablecoins. The proposed STM model would connect these workflows to delegated authority, treasury visibility and payment evidence. Availability and timing depend on the selected provider and payment rail.
THE CONTROL MODEL
A clear owner at every step.
The organisation controls its accounts, card arrangements and wallets. Its chosen bank, payment provider or custodian retains execution and the relevant payment credentials.
- YOUR ORGANISATION
Set the authority
Authorised people define the agent’s purpose, eligible counterparties, funding sources, spending limits and expiry.
- EXTERNAL AGENT
Propose the payment
The agent submits a specific purchase or payment request, with the commercial evidence needed to assess it.
- STM · PROPOSED EXTENSION
Check the instruction
Deterministic rules would validate authority, funds and limits, reserve spending capacity and obtain any required human approval.
- YOUR PAYMENT PROVIDER
Authorise and execute
The chosen bank, card payment provider or custodian would validate authority for the exact payment and execute through its supported rail.
- STM · PROPOSED EXTENSION
Verify the outcome
Independent observations would match the authorised instruction to final settlement evidence, with differences held for review.
Proposed workflow: the AI proposes a commercial action. Deterministic policy and the external payment provider would enforce the payment boundary.
DELEGATION & APPROVAL
Three outcomes for a payment request.
Approving a standing mandate sets the permitted scope. It does not mean a person has reviewed every later purchase.
ILLUSTRATIVE USD POLICY · PROPOSED MODEL
A defined spending mandate
- Automatic approval ceiling
- $25,000
- Hard per-payment limit
- $250,000
Assume a permitted purpose, approved counterparty and sufficient funds. Aggregate budgets, expiry and other checks would also apply.
A $180,000 request
Falls within the hard limit but above the automatic ceiling. The configured human approval quorum must review it.
A $300,000 request
Exceeds the mandate. It needs a separately approved mandate change or a new human-originated instruction.
Revoked authority, prohibited activity or a failed mandatory compliance check would block the request. Ordinary payment approval cannot override those restrictions.
TREASURY IN THE LOOP
Every request affects the cash position.
Count pending commitments.
The target design would reserve capacity across the agent, mandate, funding source and legal entity. Concurrent requests and retries must not spend the same budget twice. STM reservations do not lock customer funds; the external provider checks actual availability.
Recheck the exact payment.
Authority, balances and approvals can change. A short-lived decision would bind the source, recipient, asset, network and amount immediately before external signing. Changed details would require a new decision.
Distinguish status from settlement.
A bank or card authorisation, provider acceptance or transaction hash does not by itself establish final settlement. The target evidence chain would connect payment authority to bank, card or on-chain settlement evidence, using the selected rail’s settlement and reconciliation rules.
Account for payments already in flight.
Suspension would stop new STM authorisations. Outstanding signed payments, sessions and provider decisions need explicit cancellation or expiry rules, with residual exposure tracked until resolved.
APPLICATIONS TO EXPLORE
Start with a verifiable business need.
Illustrative use cases for a scoped design and pilot, with approved counterparties and a defined funding source.
Data & computing services
An agent requests an approved API dataset or cloud resource within a defined purchase purpose and budget. Recurring micropayment sessions would need separately proven provider controls.
Supplier & logistics obligations
An agent proposes a payment against an approved invoice or service obligation. Commercial evidence, beneficiary checks and the customer’s authority determine whether it can proceed.
Intercompany treasury
An agent identifies a funding need between eligible group entities. Treasury teams assess observed positions and obligations before authorising the external movement.
A PATH TO DEPLOYMENT
Prove the controls before expanding autonomy.
- 01
Design with the signing provider
Agree the use case, authority and consent model. Demonstrate exact-payment enforcement and revocation behaviour.
- 02
Test with human approval
Use a sandbox with external test signing. Exercise duplicate requests, stale data, revoked authority, outages and recovery.
- 03
Run a bounded delegated pilot
Limit the principal, counterparty, source, asset and network. Set conservative exposure limits and named release owners.
Successful reconciliation is only one test.
A payment can settle correctly and still lack authority. A pilot must demonstrate valid permission, unique execution and complete evidence, including failed and unknown outcomes.
DESIGN QUESTIONS
What needs to be established?
Can an agent change its own spending authority?
In the proposed model, an agent would submit requests and read permitted status. It would not approve payments, widen mandates, change beneficiaries or supply trusted provider evidence. Those actions need separate authority and credentials.
How would payment protocols fit?
Agent identity, checkout and payment protocols would supply inputs through separately validated integrations. The design would verify the commercial request and authority before applying the organisation’s mandate and external signing requirements. Protocol-specific integrations are proposed, not currently demonstrated STM capabilities.
Can agentic payments use fiat currency?
Yes. An agent-initiated payment can use fiat currency through a bank or card payment provider, subject to supported APIs, permissions and authentication. Stablecoins are another possible rail. STM’s agent-specific provider integrations remain proposed and require implementation and verification.
Would stablecoins make every payment instant?
Timing depends on the network, asset, counterparties, controls and any conversion or bank payout. On-chain settlement and final fiat availability are separate events.
What should a first engagement cover?
Choose one commercial workflow and a customer signing provider. Map responsibilities, implementation gaps and deployment-specific legal requirements, then agree a sandbox test plan and evidence needed for a bounded pilot.
THE NEXT CONVERSATION
Design an agentic payment workflow
around your controls.
Start with a workflow. Explore what changes with STM.