Uber dropped a sobering number into the debate over AI agents this week. During its busiest week last quarter, the company said only four people booked a ride using an AI agent. That figure lands hard against a year of predictions that software assistants would soon handle shopping, scheduling and transportation on behalf of users.
The gap between agent hype and agent use is not unique to Uber. Developers have shipped tools that can browse the web, fill forms and complete purchases, but reliability problems and trust concerns keep most consumers from handing over real world tasks. A failed restaurant reservation is annoying. A failed ride booking at midnight is a genuine problem, and users seem to know it.
Uber has experimented with AI features inside its own app, and it has partnerships that let other platforms request rides. Yet the company's own data suggests that even when the plumbing exists, demand for agent driven booking is negligible. The four rides represent a rounding error against the millions of trips Uber completes in a typical week.
The number is likely to be cited by skeptics for months. It also gives agent developers a concrete benchmark. If the category is going to justify its funding, the next step is not a better demo but a reason for ordinary riders to trust an assistant with a task they can already complete in seconds themselves.