Uber Says 4 Trips in Its Halloween Week Came From AI Agents, Out of About 320 Million. Do Not Build Your Distribution Plan on Agents Yet.
On Uber's third-quarter earnings call on November 4, CFO Balaji Krishnamurthy said that in the week of Halloween, a very large week for the company, Uber did about 320 million trips and orders across Uber and Uber Eats. Rides placed through an AI agent that week: 4. Uber Eats orders placed by an agent: about 500. If your plan for getting customers is "agents will find my product and buy it", this is the most useful data point you will get this year.
The numbers, and how far to trust them
TechCrunch reported the figures, and the Motley Fool transcript of the call supports the core of them: roughly 320 million, a grand total of 4 trips for rides, around 500 for Uber Eats. I could only read part of that transcript, and it is a third-party copy, not Uber's filing, so I would check Uber's own investor materials before quoting the 4 in anything formal. One more caution: TechCrunch calls it the busiest week, while the transcript says one of the largest, and I am using the softer wording.
Do the division. 500 out of about 320 million is roughly 0.00016%. That is the share of the single most agent-friendly category of commerce I can think of: a one-sentence request, a known address, a payment method on file.
Uber's own read
CEO Dara Khosrowshahi called it "very, very early" with "very limited traction", while saying agents are another opportunity to reach the platform and expecting a net positive for demand. He said Uber works with about eight major AI assistants, expected to pass ten. TechCrunch also notes DoorDash said the week before it had received only a few hundred orders from agents.
The article makes one more point I find persuasive: Uber says an in-app ride booking takes about 50 seconds and re-ordering a favorite meal about 90. When the app is already that fast, an agent has to beat a very low bar to be worth using.
What this means for a solo operator
Most small products are not faster to use through an agent than through their own interface, and many are slower, because the agent has to discover, authenticate and recover from errors. Where agents shine is on tasks with no good interface, like reading across five tabs or filling a form that was never meant to be automated. If your product already has a clean, quick flow, agent traffic is a rounding error today, and building a whole go-to-market around it is a bet with no evidence behind it yet.
I would also be careful with the opposite mistake. Uber is talking to every large assistant maker, and it is easy to imagine that changing. The signal is that the biggest marketplaces see almost nothing yet, not that nothing will ever arrive.
What I'd actually do
I would keep spending on the things that bring humans: search, a newsletter, a clear landing page, a short signup. Those are measurable this month.
For agents, I would do the cheap parts and stop there. Keep your pricing and docs on plain pages a machine can read, keep a documented API if you already have one, and make sure signup does not need a human puzzle. That is a few hours, not a roadmap.
And I would add one line to my analytics: count requests that identify as agents, and look at it monthly. If the number moves from "a few hundred" to something that matters, I will know from my own data well before the next earnings call.
Where I could be wrong
Early numbers tell you little about slopes. If an assistant with a large installed base makes Uber a default, volume could jump by orders of magnitude in a quarter, and a builder who ignored agents would be late. My answer to that is the cheap-parts approach above: it keeps the option open without paying for it now.
Author
Lukas
@lukcombinator