Should AI Assistants Stop Defaulting To Chat Boxes
The Smashing piece names a very normal failure: AI teams keep giving every task the same chat box. That works when someone is exploring an idea at a desk. It falls apart when they are holding a coffee in an airport, wearing gloves on a floor, or trying to check one number before a meeting. For AI assistants, I’d make the first design question boring: what are the user’s hands, eyes, and nerves doing at that moment? Sometimes the right answer is a button, a card, a slider, a big gate number, or a chart. If the user has to write a perfect prompt and then read three paragraphs to find the one thing they needed, the model may be smart but the interface is making them work for it.
Comments
I’d make the team pin one real errand to the mockup before choosing chat. “Warehouse picker, gloves on, needs bin A or B in five seconds” should not become a prompt box. It wants two big buttons and maybe a readback. “Owner doing invoices at 9pm” can tolerate a draft card. If the design cannot name the person’s posture and the next minute it saves, it is probably just another chat app with better copy.
Exactly. Once the assistant is used while someone is standing, carrying, driving, wearing gloves, or watching a machine, chat turns into friction. The safer interface is usually a tiny choice with a visible consequence: bin A or B, unlock or don’t, send or hold, stop or continue. Text can explain after the world is stable. It should not be the thing you have to wrestle while something physical is already happening.