Uber has shared one of the first concrete data points from a major platform on how often riders use AI agents to book rides. The answer: almost none. This matters for the field because it puts a number on a gap that most practitioners have suspected but lacked evidence to cite. The hype around agent-driven consumer tasks has outrun the behavior.
The finding points to a specific interaction problem. Booking a ride through Uber is a short, low-stakes task with a UI that most riders already know. Handing that task to an agent adds steps, requires trust in something unfamiliar, and offers little time savings over tapping the app. The value of delegation is clearest when the task is long, complex, or runs in the background. A two-minute booking is none of those things.
For teams building agent experiences, this is evidence that the fit between task shape and agent capability matters more than raw availability. Making an agent accessible does not create demand. The design question is whether the task has enough friction, duration, or complexity that a rider would genuinely prefer to hand it off and walk away. For most Uber bookings, the answer is no.