Uber says third-party AI agents booked only four rides in the third quarter of 2026. The Verge reported the figure from Uber's earnings call on November 4, 2026. The same report says agents placed "around 500" Uber Eats orders during the week of Halloween.
This is the first time Uber has given a count of agent-placed orders, according to The Verge. The number is tiny, but the context matters: what the base is, who said it, and what the filing does and does not show.
TL;DR: what is claimed and what is verified
| Question | Status |
|---|---|
| Third-party AI agents booked 4 rides in Q3 2026 | Reported by The Verge from the earnings call; not in the written release |
| About 500 Uber Eats orders from agents in Halloween week | Same source |
| Uber's total activity in Q3 2026 | Confirmed in the 8-K: 4,061 million trips, 59.8 billion dollars Gross Bookings |
| CEO quote that it is "very, very early" | Reported by The Verge from the call |
| Whether agents will disintermediate Uber | Open. This is one quarter and a tiny base |
| Transcript or audio checked by us | No |
What did Uber actually say?
According to The Verge, Uber executives disclosed on the call that third-party AI agents arranged four rides in Q3. They also recorded around 500 Uber Eats orders from agents in the week of Halloween. The report compares those figures with a total of 320 million Uber rides and Uber Eats deliveries.
CEO Dara Khosrowshahi is quoted as saying, "We hope that there will be more traction." He added that Uber wants services to be available to consumers in a delightful, easy way, that its direct services will be "the most delightful and easy," and that "we will work with these third parties." On the timing, The Verge quotes him: "It is just very, very early, and we want to make sure we are not dependent on any one player."
We could not retrieve a transcript, so we are relaying The Verge's quotes and figures, and we have not checked them against the audio.
What does the earnings filing confirm?
Uber's Q3 2026 results release, filed with the SEC on November 4, confirms the scale around the agent figures:
- Trips grew 16 percent year over year to 4.1 billion, or 4,061 million.
- Gross Bookings grew 20 percent to 59.8 billion dollars.
- Monthly Active Platform Consumers reached 215 million, up 14 percent.
- Uber expects about 700 fully driverless vehicles on its platform by year end and over 2,000 in 2027, according to Khosrowshahi's quote in the release.
The release does not mention third-party AI agents. So the four-ride and 500-order figures are not in the written filing we could open. They come from the call.
Is the base period clear?
Not entirely. The Verge compares four rides in a quarter and about 500 orders in one week with "320 million" transactions. The release shows 4,061 million trips in the quarter. Divided across 13 weeks, that is roughly 312 million trips a week. So the 320 million figure looks like a weekly total, which would make the 500 Eats orders roughly one in 600,000 weekly trips.
This is our arithmetic, not an Uber statement, and The Verge does not say which period the 320 million covers. Readers should treat the ratio as approximate. Either way, the share is far below one percent.
Why does this matter?
A small cart with a green price tag, showing the Uber AI agents orders that third-party agents placed on the platform
Experts have predicted that platforms like Uber and DoorDash could be hurt by disintermediation, where customers ask an AI agent to order food or book a ride instead of opening the app. The Verge names agents such as Meta's Muse and OpenAI's Dots as examples. We compared those products in our Dots vs Grok Bot vs Meta Muse post.
The Uber numbers give an early reading. Four rides in a quarter means the risk is, today, very small. The Verge adds that early agents still face friction, and that platforms have little reason to let agents replace their own customers.
Two points of caution apply:
- One quarter is a snapshot. Agent products from Meta and OpenAI launched only recently. Stripe's Link reported that agentic transactions rose 38 times in September, though only as a multiple with no base, as we covered in Stripe Link's 38x claim. A tiny base can grow fast.
- Access is a choice. Uber controls whether agents can use its platform at all. The Verge notes that getting Uber to agree to more disintermediation "will be a tough ask." The count of four rides may reflect how open Uber is, not only how much demand exists.
How does this fit the wider agent-commerce picture?
Agent commerce depends on standards for who may act for a user. We covered the open Personal Agent Protocol and how Meta Muse is becoming a transaction layer with PayPal, Shopify and Expedia. In that picture, a marketplace can be either a partner that lists inventory to agents or a gatekeeper that blocks them.
Uber says it is open to working with third parties while building its own tools. The Verge mentions the Cart Assistant for Uber Eats, which builds a grocery list from a photo of a recipe or a text prompt. That is the "direct" path Uber hopes customers keep choosing.
Uber is also a heavy user of AI inside the company. See our reports on Uber putting all 30,000 employees on ChatGPT Work and on the privacy promises of agent products, which matter when an agent acts on your behalf.
What this means for what you build
A card tapping a payment terminal with a green pulse ring, picturing an agent completing checkout for agentic commerce
If you build an agent that books or buys, the Uber number is a warning about integration, not demand. Most large marketplaces do not yet expose clean agent access. Plan for these paths:
- Official APIs or protocols first. Use a partner program or an open protocol where one exists, and avoid screen scraping, which platforms can block.
- Design for confirmations. Show price, pickup and cancellation terms before an agent commits money.
- Expect low volume at launch. If Uber saw four rides, your first integration will also be small. Measure completion and refund rates, not only attempts.
- Keep a human fallback. When an agent fails at checkout, the user should land in the app with the cart filled in.
If you run a marketplace, the lesson is the reverse. Decide now what you want agents to do, and publish rules for them. Uber's choice to "work with" third parties while pushing its own app is one template.
What are people asking about this number?
Is four rides a failure of the agents or a success for Uber? It is neither on its own. A count needs a denominator and a trend, and we have only one quarter. What the number does show is that the first consumer agents did not move Uber's totals. The release reports 20 percent Gross Bookings growth, which came from Uber's own app and partners.
Why would the Eats count be higher than the rides count? The Verge does not explain. One plausible reason, which is our guess, is that food ordering is simpler for an agent than a ride, since there is no live driver match and no pickup timing to manage. Uber has not said so.
Could Uber be hiding larger agent volume? There is no evidence of that. Uber chose to disclose the count for the first time on this call, and it did so with figures that look small. A company worried about disintermediation would have little reason to downplay a large number, but we cannot rule out that agents also reach customers through partner channels that Uber counts as its own.
What would change the picture? A named partnership that exposes Uber's inventory to an agent, a repeat count in the next quarter, or the same disclosure from DoorDash or Instacart. Until then, the safest reading is that third-party agents are still an experiment at Uber, and that Uber has chosen to keep control of the interface.
What is still unconfirmed
- The audio or transcript. We did not check the call.
- The base period. The 320 million figure's period is not stated in the report.
- Which agents. The Verge does not say which agents placed the four rides.
- Whether it was allowed. No detail on whether the rides came through a partner integration or through the app interface.
- Trend. There is no earlier quarter to compare, since this is the first disclosure.
What to watch next
The next signals are Uber's Q4 call in February, any partner announcements with the agent makers, and DoorDash's own disclosure, which The Verge links as a related debate. If a second marketplace reports a count, the comparison becomes much more useful.
Related reading
- Stripe Link: agent purchases up 38x in September, claimed vs verified
- OpenAI Dots vs Grok Bot vs Meta Muse
- Meta Muse as a transaction layer with PayPal, Shopify and Expedia
- Personal Agent Protocol: an open standard backed by Sierra, Meta, Walmart and Stripe
- AI agent privacy promises: Meta Muse vs OpenAI Dots
- Uber puts all 30,000 employees on ChatGPT Work
- The Verge report and Uber's Q3 2026 release
Details are accurate as of November 4, 2026. Agent figures are as reported by The Verge from Uber's earnings call and are not independently verified.