Agent-to-agent commerce
Two pieces of software negotiate and settle with each other, each acting under its own delegated authority. This is where the requirements above stop being optional.
Autonomous Finance
What changes when software can initiate financial actions without a human click? Most financial controls quietly assume a person is present. Autonomous finance removes that person from the moment of action, and the controls have to move into the infrastructure.
Why it changes infrastructure
A human approving a payment carries identity, judgement, a sense of limits and a memory of why. None of that is written down in the payment API. It lives in the person.
When an agent initiates the same payment, those properties do not come along automatically. An API key carries whatever authority the key was issued with. A retry loop has no instinct to stop. A budget is only a budget if something enforces it.
Agents should express financial intent. The infrastructure decides whether, and how, that intent executes.
What autonomous actors require
Each requirement replaces something a human operator used to provide implicitly. None of them is specific to AI. They apply to any software that initiates financial actions on someone's behalf.
| Requirement | With a human click | Without one |
|---|---|---|
| Identity | A logged-in person whose name is on the action. | An identity for the agent itself, distinct from the credentials of the system it runs in. |
| Authority | A role someone was given, and their judgement about using it. | Delegated authority with explicit scope: which operations, which wallets, on whose behalf. |
| Budget | A person noticing that a number looks too large. | Limits that hold across many small actions, not only per transaction. |
| Policy | Someone reading the screen before pressing the button. | Rules evaluated before execution, by the system that moves value, not by the agent. |
| Execution | A person checking whether the first attempt went through before clicking again. | Idempotent execution that tracks finality, so a retry cannot quietly become a second payment. |
| Proof | Someone remembering why they did it. | A record linking the intent, the authority used, the attempts made and the resulting ledger effect. |
| Recovery | A person who sees that something went wrong. | Explicit handling of ambiguous outcomes, with escalation to a person when evidence is insufficient. |
Example applications
Some of these are extensions of flows companies run today. One is a direction we think the infrastructure must be ready for, and is labelled as such. Agent-initiated execution in Orangepill runs through Agent Runtime, which is in Private Preview.
An agent watches wallet balances and requests rebalancing transfers within allocation policy. The runtime executes the transfers; the ledger records the movement.
Treasury & Stablecoins →Payout requests are evaluated and initiated by software against limits and approved routes, instead of being keyed in by an operations team.
Cross-Border & Remittance →An agent places an order and pays on a customer's behalf, inside a scope and budget the customer delegated.
Commerce Platforms →Two pieces of software negotiate and settle with each other, each acting under its own delegated authority. This is where the requirements above stop being optional.
In the platform
This page describes the problem. The capabilities below are where Orangepill addresses it.
Identity, scoped authority, budgets and policy for software that acts on someone's behalf.
How agents get bounded authority →Evidence of what an agent asked for, under which authority, and how the operation unfolded.
Explore Proof of Record →What happens when an agent-initiated operation ends in an ambiguous state and no person is watching.
See how Resolve works →Authoritative financial state, so automation changes balances only through recorded financial events.
Explore Wallet & Ledger →Bring the flow you want to automate. We will walk through the authority, limits and recovery it would need.