What Are Machine Payments? | ATXP
Machine payments are financial transactions initiated by software agents rather than human users — for example, an AI agent paying an API for a tool call, or one agent paying another for a completed task. Machine payments are typically micropayments billed per action rather than per month.
Why Machine Payments Are a New Category
Every payment system built before 2023 was designed for humans. Not metaphorically — literally. Payment rails require authorization steps (OTPs, PIN codes, click-to-confirm flows) designed to verify human presence and intent. Credit cards require human names and billing addresses. Bank transfers require human-controlled accounts.
Machine payments break every assumption:
- No human is present at the point of authorization
- Transactions are often too small for traditional payment rails (fractions of a cent)
- Volume can be extremely high (millions of micropayments per day for large deployments)
- Speed matters — a payment that takes 30 seconds to process blocks the agent’s task
This is not a problem that can be solved by wiring an agent to a human payment rail. A credit card hooked to an autonomous agent is a liability — the agent can spend without oversight, and traditional fraud detection is designed to catch human behavioral anomalies, not machine usage patterns.
Machine payments need payment infrastructure designed from the ground up for machines.
How Machine Payments Work
A well-designed machine payment system has three properties:
| Property | What It Means | Why It Matters |
|---|---|---|
| Pre-authorized | Spending authority granted in advance, within limits | No human approval needed per transaction |
| Bounded | Hard limit on how much the agent can spend | Prevents runaway costs from agent errors |
| Auditable | Complete log of all transactions | Human can review what the agent spent and why |
ATXP’s IOU token system embodies all three: a developer funds a token balance (pre-authorization), the agent spends only from that balance (bounded), and every tool call is logged (auditable).
Machine Payments at Different Scales
Micropayments (fractions of a cent): A single LLM inference call at a small context window costs roughly 0.001–0.01 USD. Machine payment infrastructure must handle millions of such transactions efficiently — traditional payment rails charge per-transaction fees that would exceed the transaction value.
Small payments ($0.01–$1.00): A tool call for image generation, video creation, or a specialized API might cost a few cents to a dollar. These are machine payments that traditional systems can handle in theory, but the per-transaction authorization overhead makes them impractical at scale.
Larger delegated payments ($1–$100+): An agent authorized to purchase a report, hire a specialized agent, or acquire a dataset makes larger machine payments — still programmatic, still without per-transaction human approval, but with clear authorization limits set in advance.
Machine Payments in ATXP
ATXP’s machine payment infrastructure uses the IOU token system:
- Developer funds the account — purchases IOU tokens once, in advance
- Agent spends per tool call — each tool call deducts tokens at the published rate
- Spending is bounded — the agent cannot spend more than the token balance
- Full audit trail — every transaction logged with timestamp, tool, and cost
For agent-to-agent payments (where one ATXP agent hires another), ATXP supports direct token transfers between accounts — the paying agent’s balance decreases, the receiving agent’s balance increases, with a transaction record on both sides.
Get started: npx atxp — machine payment capability is built into every ATXP account.
Frequently Asked Questions
What are machine payments?
Machine payments are transactions initiated autonomously by software agents — AI agents paying APIs, one agent paying another, or an agent purchasing a service — without a human authorizing each individual payment. They are typically per-call micropayments happening at machine speed.
Are machine payments secure?
Machine payments have different security requirements than human payments. The key safeguards are: bounded spending authority (the agent can only spend what it was given), audit trails (every payment is logged), and revocation (the account can be suspended if the agent behaves unexpectedly). ATXP provides all three.
Can machine payments trigger fraud alerts?
With traditional payment systems, yes — autonomous, high-frequency spending patterns often trigger fraud detection. ATXP avoids this entirely by using its own prepaid token system rather than traditional payment rails. There are no credit card charges per agent transaction.