Uber reveals: AI agents booked just 4 rides in a quarter out of 320 million transactions

For the past year, analysts have warned the market that AI agents such as Meta Muse or OpenAI Dots would become intermediaries between customers and services, turning Uber and DoorDash into faceless infrastructure. The first real numbers from Uber considerably cool that picture.

What exactly the company disclosed

On a call with investors, Uber's management for the first time named the volume of orders that came through third-party AI agents. In the third quarter of 2026, agents booked just four rides. Not four thousand and not four million, but four.

Food delivery fares noticeably better, but is still modest: about 500 Uber Eats orders during Halloween week. The total volume over the same period was 320 million rides and deliveries. The agents' share here rounds to zero.

Why so few

The first reason is friction. An agent has to handle addresses, clarifications, payment, item substitutions and cancellations. Each of these steps breaks more often than in the native app, which Uber has refined for years.

The second reason is incentives. CEO Dara Khosrowshahi says Uber is open to working with third parties, but its direct services will be the most convenient. Translation: the platform has no interest in letting an agent stand between it and the customer. Along with the customer go data, cross-selling, advertising and subscriptions.

The third reason is that Uber is building its own assistants. For example, Cart Assistant in Uber Eats assembles a grocery list from a photo of a recipe or a text request. It is the same "tell us what you want" scenario, but inside the Uber ecosystem.

What this means for developers and product managers

If you build agents that order food, taxis or tickets, the main risk is not model quality but access. Without an official API or partnership agreement, you depend on an interface that the platform owner can change or shut down. Plan agreements with platforms as seriously as you plan the technical side.

If you own a marketplace or service, Uber's numbers point to a sensible strategy: don't close off completely, but don't hand over your customers either. Open limited agent access to see demand and avoid falling behind, while strengthening your own assistant in parallel.

Limitations

This is one quarter and one company, and the agent market is only taking shape. Khosrowshahi himself calls the situation "very, very early" and admits he would like more traffic. A zero share today does not guarantee a zero share in two years if agents solve the friction problem and platforms agree to deeper integration.

For now the conclusion is simple: AI agents as a distribution channel for everyday-demand services remain more of a hypothesis than a reality. Those building businesses on intermediary agents should build their models on figures like four rides, not on analysts' optimistic forecasts. Those defending their platforms should track the trend every quarter: that is what will show when the fear becomes justified.