AI Product Interface Study
Interface study, not a product.
Objective
Explore what an AI product feels like when the interaction model is not conversation.
Process
Rebuilt one assistant flow three ways — inline suggestions, a command palette, and an ambient rail — then compared them on interruption cost.
Prompt
Design an AI surface with no chat input anywhere in it.
Result
The ambient rail was least intrusive and most ignored; the command palette was fastest for people who already knew what they wanted.
Learning
Interaction model should follow expertise level — I would ship both, not pick one.