AI agents barely register at Uber so far. On the company's third quarter earnings call on 4 November, chief financial officer Balaji Krishnamurthy said Uber and Uber Eats handled about 320 million trips and orders in the week of Halloween, one of its busiest weeks. Around 500 of the Uber Eats orders came from AI agents, and agents booked a total of 4 rides. It is the first time Uber has put a number on agent traffic, and it is one of the few real usage figures in a debate that is mostly forecasts.
What Uber said on the call
The numbers came in answer to an analyst's question about AI agents. According to the call transcript published by The Motley Fool, Krishnamurthy said the week before the call was one of Uber's largest because of Halloween, with just around 320 million trips on Uber and Uber Eats, around 500 Uber Eats orders from AI agents and a grand total of 4 rides fulfilled from agents. Chief executive Dara Khosrowshahi added that it is very early, with very limited traction so far.
Reports differ on the period. The Verge's headline describes the 4 rides as last quarter, while the transcript and TechCrunch tie the figures to the Halloween week. We follow the transcript, which is also consistent with the calendar: Halloween fell on 31 October, after the third quarter ended on 30 September. For scale, Uber reported 4.1 billion trips across the whole third quarter, up 16% on a year earlier.
Why the agent number is so small
The worry for apps like Uber and DoorDash is disintermediation, where people stop opening the app and ask an assistant such as ChatGPT, Gemini or Claude to order for them, and the assistant owns the customer relationship. Khosrowshahi argued that Uber's own app is already hard to beat on speed. He said the average ride booking takes 50 seconds, 75% of rides are booked with a single tap based on personalised destinations, the median time to reorder a favourite meal is 90 seconds, and members now account for more than half of Uber's bookings. TechCrunch notes that this suggests an agent is not necessarily saving anyone time.
Uber is not alone. TechCrunch reports that DoorDash said on its own call the week before that it had received only a few hundred orders from AI agents so far. Both figures are small next to the attention agent shopping has received this year. TechCrunch points to Meta's Muse reaching the top of Apple's US App Store last month with more than 10 million users, many of them shopping through it, according to Meta.
What Uber is doing about agents
Uber is working on two fronts. It is building its own assistants, including a Cart Assistant for Uber Eats, which The Verge says builds a grocery list from a photo of a recipe or a text prompt. Khosrowshahi said early results show baskets built with it are roughly twice the size of an average basket. It is also talking to the AI companies. He said every AI system company is talking to Uber, that it is negotiating the terms of trade, and that he counts eight assistants today and expects more than ten. He described third-party agents as another way to reach Uber's platform and a likely net positive for demand.
His caution is personal. Khosrowshahi said he used to work in the travel business, where companies depended on Google for customers, and that he does not want to go back there. That explains why Uber wants agents to be one channel among many rather than the front door.
What the number does and does not show
The figure is a snapshot of one week, given verbally on a call, and it covers agents that placed orders through Uber, not people who used an assistant to research a trip and then opened the app. It also comes early in agent adoption, so it says little about where volumes will be in a year. What it does show is that, today, a fast app with saved preferences still wins easily for repeat purchases, and that companies with strong apps have time to set terms with AI providers.
We think this is one of the most useful agent statistics of the year so far, because it measures what customers actually did, and it should temper claims that agents have already taken over consumer ordering.
What to do this week
- If you sell to consumers, measure how long a repeat purchase takes in your own app or site and remove steps where you can.
- Track orders that arrive from AI agents as a separate channel, even if the count is zero today.
- When an AI company asks for access to your catalogue or booking flow, read the data and pricing terms carefully before you agree.
- Once a month, try buying from your own business through a major assistant and note where it fails.
For a related example of an agent built for store owners, read our piece on Perplexity Computer for e-commerce, and see Stacks for tool combinations that work for small teams.