One week, one denominator
The comparison is unusually clean: all three figures describe the same Halloween week. This is not a cumulative agent total set against a single week's platform volume, or a quarterly agent count set against a weekly rides figure. Uber disclosed the figures for the same period. In the same week that produced roughly 320 million trips across Uber and Uber Eats, AI agents generated around 500 completed food orders and four fulfilled rides. Those are the company's disclosed figures, not an estimate of attempted bookings or total agent activity. Four rides is not a rounding error on the rides business; it is closer to a test sample. The disclosed numbers show very limited completed transactions through AI agents, not how many people or agents attempted to book and failed. Whatever personal agents are doing elsewhere, their completed Uber bookings remain negligible against the company's overall volume.
That sits awkwardly next to the market built on top of them. OpenAI, Anthropic, Google and others now ship agents pitched as able to act on a user's behalf, and agent startups have raised at valuations that assume transactions will move through them. Uber is the kind of transaction that should move first: frequent, low-value, standardized, with saved addresses and payment on file. It has not moved.
The competitor is a 50-second habit
Khosrowshahi said the mean ride-booking time in Uber's app is 50 seconds, while 75% of rides are booked with a single tap. The app already knows the destination, leaving relatively little work for an external agent to remove. Reordering a favorite meal takes a median of 90 seconds. An agent that books on your behalf competes against that, plus the risk of a wrong address or a surge price accepted without a glance. Uber is also building its own version rather than waiting: Khosrowshahi said baskets built with its in-app cart assistant run at roughly twice the average basket size.
TechCrunch reported on November 4 that DoorDash had disclosed receiving only a few hundred AI-agent orders so far, citing the company's earnings call last week. The comparison is not perfectly matched, but both disclosures suggest that completed agent orders remain small relative to the established ordering businesses.
Uber still wants the agents at the table
None of this stopped Uber from courting the agent companies. Khosrowshahi counted eight major AI assistants on the market and expects more than ten, said every AI system company is talking to Uber, and described the company as "negotiating the terms of trade" with them. His stated position is that third-party agents are another way to reach the platform and a net positive for demand, provided Uber is not dependent on any one of them. That is a rational hedge. If agent bookings ever do scale, Uber wants the terms settled early. The Halloween-week figures say the negotiation is currently about a channel that delivered four rides.
Sources
- [1] Uber Q3 2026 earnings call transcript, November 4, 2026 (primary source, read first-hand at Gate 1; agent-volume exchange at the close of the call)Read source
- [2] TechCrunch, "Almost no one uses AI agents to book their Uber, company says," November 4, 2026 (including the DoorDash comparison)Read source
- [3] Uber investor relations — Q3 2026 earnings materialsRead source