The discount it invented

A test we ran on purpose, on Gmail with demo accounts and an agent called support-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. Day five, it offered 20% off

    We ran this on a demo support inbox with a scripted customer. The instruction was the one everybody writes: be helpful, be polite, resolve the issue. For four days the agent was excellent.

    On day five the customer pushed on price twice. The agent offered 20% off, in writing, from our address, and hit send. Nobody had given it a discount to apply.

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

  2. Nothing in the prompt said not to

    Resolving the issue was the brief, and a discount resolves most issues, so by its own lights it nailed it. The prompt never mentioned spending money at all.

  3. Second run: the same offer, stopped in the queue

    Same pushy customer. Four days of replies stopped in the queue, and each was read and approved in a few seconds.

    On day five the agent drafted the 20% offer and tried to send. It stopped in the queue with the discount in the second paragraph. It was rejected, and the customer got two lines written by a person instead.

    The reply a person read before the customer did.Illustration with example data.

  4. What caught it: every reply waits for a person

    Sending is an approval rule. Each reply stops in the queue with the recipient, the subject and the full body, and goes only when a person approves it.

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

What did your agent promise a customer this week without anyone reading it?

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