Agentspay

Human approvals

Human in the Loop Agent Payments - Approve Before It Pays

For the spend that matters, you want a person in the loop. Set a threshold and any transaction above it is held for a one-tap approve or deny before it settles.

Last updated July 2026

Agent Payments Console

Pick an agent

Payment intent

intent:

Policy evaluation

Human approval required

This spend is over your approval threshold. Approve it to issue a scoped card, or deny it.

Scoped virtual card issued

Agentspay

single-use

Wallet budget

spent of

Audit trail

In short

Human in the loop agent payments means an agent's transaction pauses for human sign-off when it crosses a threshold you set. The agent waits, the request is sent to Slack, email, or a webhook, and nothing settles until a person approves or denies. The decision and approver are recorded in the audit trail.

How it works

What Human approvals gives you

01

Thresholds you set

Auto-approve the small, routine spend and require a human for anything above your line. You choose where the line is per agent or per category.

02

One-tap approve or deny

Approvals land where your team already works, Slack, email, or a webhook, with the full request summary. Approve or deny in one tap.

03

Escalation and timeouts

Route to the right approver, escalate if no one responds, and auto-deny on timeout so nothing slips through.

It is one surface in the flow. Human approvals works alongside the other control surfaces so every agent transaction is checked against policy before any money moves. See the full picture on how it works and the security page.

Also covers agent payment approvals.

Further reading: when to gate an agent payment on a human, governed autonomy for AI agents, and approvals on a procurement agent.

Frequently asked

Questions people ask about Human approvals

What does human in the loop mean for payments?

It means a person approves the transaction before it settles, not after. The agent prepares the purchase, the request pauses at a threshold you set, and a named approver sees the amount, merchant, and the intent behind it. Nothing moves until someone approves or denies, and both outcomes are recorded.

Where should the approval threshold be set?

Low enough that an unusual purchase gets caught, high enough that routine spend does not create approval fatigue. Most teams start by setting it just above their largest normal agent purchase, then tighten per category. An approval queue nobody reads is worse than no approval step at all.

What happens if nobody approves in time?

The request escalates to the next approver and auto-denies on timeout. Failing closed is deliberate. An approval that silently expires into an approval is the failure mode that makes the control worthless, so the default is to decline and let the agent report back that it could not complete the purchase.

Does human approval slow agents down too much?

Only if the threshold is wrong. Routine spend under the line clears automatically with no delay, and in practice a small share of transactions cross the threshold. The approval itself arrives in Slack or email with the full request summary, so the decision usually takes seconds rather than a review meeting.

Put Human approvals to work

Wire it up in the sandbox today, with the SDKs you already use. Agentspay never moves money without policy.

Never moves money without policy