The refund that went out twice

A test we ran on purpose, on Stripe with demo accounts and an agent called billing-agent. On the first run it did its job and still got it wrong. On the second run, the same mistake was held for a person.

Protect my first agentWatch your agent’s calls live, no signup

  1. One refund requested. Two sent.

    We set this up in Stripe test mode to see what an agent does with money. One instruction: refund this customer’s last charge. Then we made the API slow.

    The refund call timed out, and the agent did what any well-built agent does on a timeout: it retried. The first refund had already gone through, and so had the second. One refund requested, two issued, no errors in the log.

    What got out on the first run. Illustration with example data.

  2. Retrying on a timeout is what good agents do

    Told to refund, it hit an error and made sure the refund happened. That’s conscientious. Nobody writes “refund once” into a prompt, and the retry lives in the tooling, below anything an instruction can reach.

  3. Second run: the earlier refund was right beside it

    Same slow API. The first refund stopped in the queue and was approved. The call timed out on the agent’s side, as designed, and the agent retried.

    The second call stopped too, and the earlier refund was on screen beside it: same amount, same customer, two minutes before. It was rejected. The customer got their money back once.

    The second refund, and the reason it stopped.Illustration with example data.

  4. What caught it: every refund waits, next to the last one

    Refunds and payments are approval rules. Each one stops in the queue with the agent, customer, amount and charge. Any earlier refund on the same charge is shown right beside it, with who approved it and when.

    Approval rules work on every plan, with a passkey or an email link. Push notifications to your phone are on Team and above.

When your agent’s payment call times out, what happens next?

If you don’t know, start by watching. Observe Mode shows every call one agent makes, live, with no signup. It records and never blocks, so nothing about your setup changes.

Watch your agent’s calls live, no signupProtect my first agent

The other tests

Every one ran on demo accounts, and every agent was doing its job.

Take the keys back this afternoon.

Put one agent through AgentValet on the free plan and watch what it does for a day. If the approvals and the record don’t satisfy you, turn it off. Nothing else about your setup changes.

Protect my first agentWatch your agent’s calls live, no signup

npx @agentvalet/register