Uber says only four rides and 500 Uber Eats orders came via AI agents in its busiest October week

On Uber's Q3 2026 earnings call on Wednesday 4 November, executives said only four rides and about 500 Uber Eats orders came through AI agents during the week of Halloween. That was out of roughly 320 million trips and orders in the same week. Executives called traction very early and limited. There was no product or policy change, and the figures are reported second-hand by TechCrunch, with no transcript linked.

This is a rare hard number from a public company on real agent-driven purchasing, and it is tiny: roughly 504 orders in 320 million, far below one in a million. It backs up DoorDash's similar figures (a few hundred agent orders). Anyone sizing the market for agent commerce, or deciding whether to build checkout and booking integrations for agents, should treat near-term volume as negligible. It says little about the longer term, because a single week at an early stage is a weak guide to future adoption.

Headlines like "almost no one uses AI agents" can read as a verdict on the whole agentic web, but this is one week, two companies and two product categories. We also don't know how Uber counts an "agent order", how easy it has made agent access, or whether it has promoted the feature. Low usage may reflect thin integration effort as much as weak demand.

1. 01Uber's Q4 2026 earnings call (likely February 2027): whether it repeats an agent-order figure, and whether the count moves from hundreds into the thousands or more.
2. 02Whether DoorDash, Instacart, Lyft or Booking disclose agent-driven order counts, and in what units, to allow a like-for-like comparison.
3. 03Any new integrations or promotion from Uber with OpenAI, Anthropic or Google agents, and whether agent orders then rise after launch, which would show the gap is about distribution rather than demand.