I say please and thank you to the AI. Every time. I know it doesn’t care and I do it anyway, half force of habit, half quiet insurance in case the robots rise up and come looking for the rude ones. Spare me, I was lovely (and definitely not because watching Terminator as a kid made me very aware of the Skynet possibilities).
So last week’s news made me grin. Anthropic updated its rules to ban “sustained and needless abusive or cruel behaviour” toward Claude. Not normal frustration, not pushback, not dark creative stuff. Just the extreme cases, people being cruel to the model over and over for no reason. They’ve basically given their product the right to walk away from you.
You can find that touching or daft depending on your mood, and there’s a whole noisy argument about whether Claude is “conscious” that I’ll leave to people with more time than me.
Because if you’ve put AI anywhere near your customers, the interesting bit has nothing to do with machine feelings.
Your support bot, the assistant on your product page, the thing answering people at 2am so you don’t have to, it’s built on a model you don’t own. And that model runs on rules written by the company that made it. Last week those rules changed. Nobody asked you. One day your bot takes whatever a furious customer throws at it, the next it can disengage when things turn abusive.
This particular change is harmless, good, even. Someone screaming at your support bot wasn’t a sale you were going to rescue. I’m not bothered about the rule.
I’m bothered about what it shows. You’ve built something your customers talk to, and you don’t control how it behaves at the edges. Those limits get set in a room you’ll never sit in. Today it’s “be civil to the AI.” Tomorrow it might be a topic your bot won’t touch, a tone it won’t take, a request it quietly starts refusing halfway through a flow you spent months building. And you’ll find out the way everyone does, when a customer complains or the numbers dip.
That’s the thing nobody puts in the sales deck. AI feels like software you bought. It behaves more like someone you rent, who answers to an employer that isn’t you.
Doesn’t mean don’t use it. We build on these models daily, I’d be a hypocrite to say otherwise. It means know what you’ve got. Read the usage policy of whatever model sits behind your customer experience, once, so a change doesn’t blindside you. Keep a human reachable for when the bot taps out. And assume the rules will move again, because they will.
Be kind to the AI. I am. Just don’t mistake being allowed to use it for being in charge of it.