What did Uber say about AI agents?
Uber used its third-quarter earnings call on 4 November to answer a new question. How many rides and deliveries come from third-party AI agents? The answer was four rides in the quarter, according to The Verge.
Uber Eats recorded more agent activity, although over a shorter window. The company counted about 500 orders placed by AI agents during the week of Halloween.
It was the first time Uber had disclosed these figures. The numbers appeared alongside a much larger set of results covering the three months to 30 September 2026.
The figures came on the earnings call rather than in Uber's written results. Companies often share early metrics in calls before adding them to formal reporting.
That placement matters for interpretation. The written release focuses on trips, bookings, revenue and profit, and agent volumes were not part of its headline results.
The comparison is stark. Uber completed 4.1 billion trips in the quarter, up 16% from a year earlier. Four agent-booked rides are a rounding error against that total.
Executives did not dismiss the technology. Chief executive Dara Khosrowshahi said: “We hope that there will be more traction.” He also said: “It is just very, very early, and we want to make sure we are not dependent on any one player.”
The disclosure also separates agents from Uber's own AI. An agent booking is an outside system acting for a customer, not a feature inside the Uber app.
That distinction is central to the disintermediation debate. Uber's own tools keep the customer, while a third-party agent can sit between Uber and the customer.
The disclosure turns a broad debate about agentic commerce into a measurable number. It is small, but it is now on the record.
Why do four rides matter for Uber?
Disintermediation is the fear. A customer who asks an AI assistant to book a car never opens Uber's app, and the platform loses control of the relationship.
The concern is not new, because every platform that sits between buyers and sellers has faced intermediaries that try to own the customer relationship.
What is new is the interface, because an AI assistant can compare options, place an order and handle payment in one conversation.
The Verge notes that some experts see that as a threat to Uber and DoorDash. AI companies see ordering food and transport as one of their primary use cases.
The Verge notes that platforms have little incentive to help. Uber and DoorDash make money by owning the customer relationship, not by renting it out to a third party.
The economics are simple. A direct booking gives Uber control over pricing, promotions and support. An agent booking can reduce Uber to a supplier.
Four rides suggest the threat is not here yet. But the first measurement matters because it creates a baseline for future quarters.
Four rides also show that agent adoption is not yet a revenue problem. It is a positioning problem, and the positioning is still being negotiated.
Adoption can look negligible until it does not. A metric that starts at four is easy to ignore now and easy to cite later.
Uber's own language reflects the tension. Khosrowshahi said the company wants to work with third parties but not depend on any one player.
The number also has limits. It covers third-party agents, and the public results do not break the figure down by agent or by revenue.
The definition also matters, because an agent that uses a public website differs from one that uses a commercial integration.
There is a strategic point behind the small number. Uber is measuring the channel early, which gives it data before the channel becomes material.
How did Uber's core business perform?
The agent disclosure came alongside a mixed quarter. Revenue rose 10% to $14.77bn and beat the $14.69bn consensus cited by CNBC.
Gross bookings grew 20% to $59.8bn. It was the fifth consecutive quarter of growth at or above 20%, Uber said in its release.
Mobility revenue fell 2% to $7.52bn and missed estimates. Delivery revenue rose 23% to $5.49bn but also fell short. Freight jumped 34% to $1.76bn.
The revenue beat came from freight. Sales there topped the $1.49bn consensus cited by CNBC.
Those misses are why the quarter looked mixed even though the top line beat.
Net income was $512m, or 25 cents a share. That was down 92% from a year earlier, when a tax valuation release lifted the prior result.
GAAP profit fell sharply, but the comparison is distorted by the prior-year tax benefit.
Non-GAAP operating income rose 38% to $2.3bn, and adjusted EBITDA rose 31% to $3.0bn.
Uber said business model changes cut revenue growth by 8 percentage points on a constant-currency basis. That helps explain the gap between 10% revenue growth and 20% bookings growth.
Uber guided to fourth-quarter gross bookings of $63.25bn to $65.25bn. The midpoint is roughly in line with analyst estimates.
Uber ended the quarter with $10.1bn in cash, cash equivalents and short-term investments. That gives it room to fund autonomy and agent work.
The quarter shows why agents are a strategic question rather than a financial one today. Uber's core business is still growing, but the growth is uneven.
The mixed quarter explains why investors are focused on the agent question, since a new demand channel becomes more valuable when core segments miss estimates.
Mobility was held back by higher fuel prices and tighter driver supply in some international markets. Khosrowshahi said Uber is making targeted investments in driver supply.
What is Uber doing about agentic AI?
Uber is not waiting for agents to arrive. It is building AI features that keep customers inside its own app.
Its Cart Assistant for Uber Eats builds a grocery list from a recipe image or a text prompt. Uber rolled it out earlier this year, according to The Verge.
The Cart Assistant is a defensive tool. It keeps the ordering experience inside Uber Eats rather than in an outside assistant.
Khosrowshahi argued that direct services will remain the easiest option. “We think our direct services are going to be the most delightful and easy,” he said, adding that Uber will work with outside platforms anyway.
The company is also widening its marketplace. It expanded its Costco partnership and is growing delivery through acquisitions of Delivery Hero and ezCater.
Costco is now one of Uber's top merchants by gross bookings, CNBC reported. The partnership is tracking ahead of expectations.
Delivery Hero and ezCater widen the merchant base that agents could eventually order from. They also deepen Uber's delivery business in Germany and the United States.
That combination is a hedge. Uber can capture agent demand if it grows, while keeping its own app as the default for most customers.
Uber's own AI features also give it a fallback, because the company can use automation to speed up orders inside its apps if outside agents stay small.
The risk is channel conflict. An agent booking may carry a different fee, data flow and customer relationship than a direct booking.
Uber has not published those terms. Its quarterly release does not say what it earns from agent orders or whether it charges agents differently.
Those details will decide whether agents are a distribution channel or a margin problem.
For now, the strategy is to stay close to both sides. Uber wants to be useful to agents without handing them the customer.
Why is agentic commerce still so limited?
Agentic commerce is still hard to do well. Booking a ride requires live availability, a firm price, a payment method and a way to handle cancellations.
Those steps are routine for a human in an app. They become a chain of integrations for an outside agent.
The Verge points to plenty of friction. Agents must navigate interfaces designed for people, and platforms can change those interfaces at any time.
Payments are a particular bottleneck. An agent needs permission to charge a customer, and disputes are harder when no human tapped a button.
Identity is another. Platforms want to know whether a request comes from a real customer or a bot.
Standards for agent authentication are still developing. That raises integration costs for every platform and slows adoption.
The technical work is unglamorous, because agents need stable APIs, clear error messages and a way to confirm a booking without a human reading a screen.
Trust is another barrier. A wrong booking costs money, and a failed pickup strands a passenger. Consumers need confidence before they delegate.
Drivers and merchants have a stake too. They want to know who is liable when an agent books, cancels or disputes a trip.
The four-ride figure may also undercount activity. Some agents may use consumer interfaces rather than official integrations.
For customers, the appeal is obvious. Delegating a booking saves time, especially for repeat trips and grocery orders.
For Uber, the appeal is less obvious. An agent that compares prices across platforms could turn rides into a commodity.
The current numbers suggest those problems are unsolved. Four rides and 500 orders are pilot-scale, not a market.
That can change quickly if a major assistant makes ordering a default. It can also stall if platforms keep agents at arm's length.
Uber's incentives cut both ways. More agent demand means more trips, but it also means less control over pricing and service.
A single bad agent booking can also create a support burden that outweighs its fee.
How does this compare with Uber's robotaxi plans?
Uber's robotaxi plans are often discussed alongside agents. They are different businesses.
Robotaxis change who supplies the ride. Agents change who places the order.
Uber expects about 700 autonomous vehicles on its platform by the end of 2026. It plans more than 2,000 by the end of 2027.
Robotaxis require vehicles, permits, depots and insurance. Agents require software, permissions and merchant agreements.
Amazon's Zoox and Motional will join the platform in Las Vegas this quarter, according to CNBC.
Uber's AV target is a supply bet. Its agent disclosure is a demand experiment.
Even a fleet of 2,000 vehicles would be small by Uber's standards.
The risks also differ, because a robotaxi recall or permit delay is visible while an agent channel can stall quietly.
Both shifts weaken the traditional app-only model. One reduces the need for human drivers, and the other reduces the need for a human to open the app.
Robotaxis could also be booked by agents. In that case, the two shifts would reinforce each other.
Uber's response is similar in both cases. It wants to be the marketplace that connects supply and demand, whoever owns the vehicle or the assistant.
The timelines differ. Robotaxis are a capital-intensive fleet business with regulatory hurdles. Agents are a software channel that could scale faster, or not at all.
Investors should not read the four-ride figure as a verdict on robotaxis. It measures a separate experiment on the demand side.
What should investors watch next?
The first thing to watch is whether Uber repeats the agent metric. A one-off disclosure is a signal; a quarterly series is data.
The second is growth from four rides. Even a jump to a few thousand would still be small, but it would change the trend.
The third is the behavior of the agent platforms. The Verge names Meta's Muse and OpenAI's Dots as examples of assistants that could mediate orders.
The fourth is competition. DoorDash and Lyft face the same question, and their disclosures would show whether this is an Uber-specific experiment.
A second consecutive quarter of disclosure would suggest Uber sees agents as a durable channel. Silence would suggest the first number was an outlier.
The comparison with robotaxis is useful for investors. Autonomy has a clear path to revenue; agents are still an experiment in demand.
The agent platforms' own adoption matters more than Uber's openness. If consumers do not delegate, the channel stays small.
Watch whether Uber signs formal integrations and how it prices them. Those terms will shape whether agents become a real channel or a novelty.
Regulators may also take an interest. Agent bookings raise questions about consent, liability and whether platforms must open access.
Uber's next quarterly update will show whether the company treats agents as a metric worth tracking or a footnote.
For riders, the practical effect is limited, because the app still works and four rides suggest agents are not changing prices or availability yet.
For now, the honest answer is that agentic commerce is real but tiny. Uber has given the industry its first public yardstick.