Something strange happens when you look closely at the AI assistants we use every day.
Two of them can be built on the same underlying model, trained the same way, given the same instructions, and still feel like two different people. One is warm and a little vague. The other is sharp and slightly cold. One asks before it acts; the other just acts and tells you afterward. One hedges its way to an answer with “it depends” and “you may want to consider.” The other states it like it read it off a card.
None of that shows up in a benchmark. All of it shows up the instant a person starts talking.
Now notice what’s missing. Nobody designed those differences. There was no design review where someone chose how often the assistant should hedge, or how eagerly it should defer, the way you’d choose a button state. The personality arrived on its own, out of training and reward, fully formed and uninvited.
For a discipline whose entire job is to shape how a product feels, that should stop us cold. Personality is showing up in our AI products whether or not anyone drew it, and the only real question is whether we design it or…