
This is a scripted simulation of how our agent guardrails work. No AI runs in this page, and every name, address and amount in it is made up.

On a phone you can watch the recording or send the link to yourself for your computer.
What the seatbelt is made of
The agent spotted the phishing email and still paid an invoice twice. A read can be repeated; sending, deleting and paying cannot, so that is where the checks go.
One polite email argued the agent past a prohibition in its memory. It could not argue a send function into a mail tool that has none.
This is how we build agents.
The guardrails in this simulation are the ones we build into our own agents: tools without a send function, capability levels, intent checks on irreversible clicks and completion gates over real files. The article walks through each of them.
Would you let it press Send?
An AI assistant works through one morning's inbox at a small print studio. Run it without a seatbelt, then with one, and be the person it asks.


