Pro tip: our chatbot went off script on a live demo in Austin and I had to decide in 10 seconds
Last month at a small AI meetup in Austin, our support bot demo started answering a customer question about refunds with a made up policy, right in front of about 40 people. I could either kill the demo or let it run, and I let it run while typing corrections live, which maybe was the wrong call. Half the room said I should have hard coded guardrails on that topic first, the other half said seeing it mess up was the most honest part of the night. Where do you all land on this, do you demo raw or lock it down?
That "mad up refund policy" bit reminds me of a buddy who did a live cooking demo and forgot the sugar, so he just called it a "savory version" and kept going. Letting it ride is usually the better show, since a bot that never messes up is a bot nobody trusts.