Uber only fulfilled 4 rides from AI agents last quarter

On its Q3 2026 earnings call, Uber disclosed for the first time how many trips third-party AI agents actually completed on its platform: four rides in the quarter, and around 500 Uber Eats orders during Halloween week, against 320 million combined rides and deliveries. The Verge frames this against rising talk that agents from Meta, OpenAI, and others could cut platforms out of the customer relationship.

CEO Dara Khosrowshahi said the company hopes for more traction and will work with third parties, while betting that Uber's own direct experience stays "the most delightful and easy." Uber is also shipping first-party AI — including a Cart Assistant for Uber Eats that builds grocery lists from a recipe image or text prompt — designed to keep demand inside its app.

We're not surprised by the near-zero numbers, and we don't think they settle the debate. Four rides is a product-readiness score, not a strategic all-clear. Today's agents still hit friction at login, payment, real-time ETAs, cancellations, and edge-case customer support — exactly the surfaces platforms control. Uber has little incentive to smooth those surfaces for a bot that wants to own the customer.

The interesting posture is the dual track. Publicly: openness to third-party agents, no dependency on any one player. Privately in product: invest in Uber-native AI so the assistant lives where the margin lives. That's the rational incumbent play. Disintermediation only scales if the platform cooperates or if agents reliably drive the same flows through brittle UI automation — and UI automation breaks every redesign.

I'm watching incentives more than demos. Food and ride marketplaces make money on frequency, fees, and ads inside their own surfaces. An agent that price-compares across Uber, DoorDash, and a local courier is existential if it works. An agent that books four rides a quarter is a press-cycle footnote. Khosrowshahi's "we hope that there will be more traction" is hospitality with a leash.

Useful AI would genuinely make ordering and dispatching less tedious. Eyes open: the same capability that helps a traveler who doesn't want to open five apps also threatens the apps' pricing power and loyalty loops. Platforms will productize assistants before they willingly become pipes.

If agent commerce is the AI labs' path to profit, Uber just published the baseline. The next print — tens, thousands, or still a handful — is the real signal of whether this use case is waking up or stuck in keynote land.

Concerns about AI-agent disintermediation have followed the rise of consumer agents that can browse, fill carts, and call APIs on a user's behalf. Marketplace companies have generally responded by opening limited partner surfaces while accelerating in-app assistants that preserve the primary customer relationship.