Four Rides: What Uber's Agent Numbers Actually Tell Us

Uber says outside AI agents arranged four rides in a quarter. What that small number means for Indian businesses planning to sell through agents.

A very small number, and a useful one

If an AI agent can book a ride, order dinner and plan a trip, why would anyone still open an app? It is a fair question for any founder who sells through an app or a website.

Uber has now given us a first piece of evidence. According to a report from Let's Data Science, which cites The Verge's account of Uber's earnings call, third-party AI agents arranged four rides in the third quarter of 2026. Agents also placed around 500 Uber Eats orders during the week of Halloween. Uber's total over a period the article does not specify was 320 million rides and deliveries. The Verge notes that this is the first time Uber has disclosed how many rides or deliveries outside agents arrange.

The same brief that tracks agent news this week points to a Fortune report saying Uber, DoorDash and Airbnb all saw negligible usage from agents.

Four is not a typo. It is also not a verdict.

Why it matters if you run a business

Many of us have heard the prediction that agents will sit between customers and businesses. The customer says what they want, the agent picks a supplier, and the supplier's own app or website matters less.

If that were happening quickly, the numbers would show it. They show the opposite, at least for now. For an owner deciding where to spend a quarter's engineering time, that is useful: a polished agent integration may be a smaller priority than a faster checkout, a better catalogue or a clearer price.

It also helps to read the numbers carefully. The ride figure covers a quarter and the food figure covers a single week, so they are not a like-for-like rate. They come from one company's disclosure. The article does not say how Uber defines an order arranged by an agent.

The idea in layers

Think about what an agent needs to complete a purchase on someone's behalf.

The interface. Apps and websites were built for people. An agent has to read them, or use an interface made for software, and The Verge says agents still face many friction points.

The platform's incentives. The Verge argues that platforms such as Uber and DoorDash have little reason to let an agent replace their own relationship with the customer. Uber's chief executive, Dara Khosrowshahi, said he hopes there will be more traction, called it "very, very early," and said Uber wants to avoid depending on any one player. He also said Uber will work with these third parties.

The platform's own agents. Uber is building AI tools inside its app, including one that builds a grocery list from a recipe image or a text prompt. So the contest may not be agents against apps. It may be outside agents against the platform's own agent.

The customer. Habits are slow to change. Someone who has ordered the same way for years needs a clear reason to hand the task over.

An older pattern may fit here. In earlier technology waves, businesses often rushed to build for a new channel before their customers had a reason to use it, and the habit took longer to form than the first headlines suggested. Agents could follow a similar path. That is a reading, not a finding.

What we do not know yet

We do not know how Uber counts an agent-arranged order, or whether the weekly Eats figure is typical. We do not know whether adoption will grow steadily or jump when a large assistant makes agent ordering the default. Some experts, The Verge notes, expect customers to order through assistants such as Meta's Muse or OpenAI's Dots instead of opening the app, though it judges the risk very small on these numbers.

We also do not know how Indian platforms compare. Nothing in these sources speaks to India.

Three things to try this week

  1. Measure before you build. Look at your own orders. Can you tell which ones came through an automated tool? If you cannot, add a simple tag, so that in three months you have a number of your own and not someone else's headline.
  2. Make your catalogue readable by software. Clear product names, prices, stock status and delivery times help an agent, and they also help your website, your search engine and your own team. This work pays off either way.
  3. Test one agent flow end to end. Pick a common task, such as reordering a usual item. Try it through an assistant, write down where it breaks, and decide whether the fix is worth doing now.

A calm way to read the signal

Early numbers are small for most technologies. What matters is what you learn while they are small. Four rides tell us that customers are not yet handing their errands to agents. They do not tell us they never will.

By Nikhil Teltumbade, Director, Godrix AI Solutions.
Working on something similar? Godrix AI Solutions builds AI automation and apps for business workflows. Book a free audit.

← All posts