Four rides in its busiest week: Uber shows AI agents can't yet compete with a 50-second app

By DiarioBitcoin (with coverage from TechCrunch and El Ecosistema Startup) · November 4, 2026.

Uber logged just four rides booked through AI agents during Halloween week, its busiest week in October. On Uber Eats, orders placed by agents came to around 500. The figures were presented by the company itself on its third-quarter earnings call.

For scale: CFO Balaji Krishnamurthy told analysts that Uber processed about 320 million rides and orders across both services that week. The four rides and 500 orders are distinct activities and should not be added together as a single measure, but in both cases the scale is minuscule relative to the total.

CEO Dara Khosrowshahi called the traction "very, very limited" and said he expects it to grow. His main explanation is practical: according to him, booking a ride in the Uber app takes 50 seconds on average, and reordering a favorite on Uber Eats about 90. With processes that short, an external assistant barely saves any time.

It is not an isolated case. According to the article, DoorDash said the previous week that it had received only a few hundred orders from AI agents, against more than 3 million deliveries a day, and that most of its customers were still ordering from its app.

The backdrop is a lot of noise around personal agents. The source mentions Instinct, a startup valued at $10 billion, and Meta's Muse, which according to a Meta report cited by TechCrunch has surpassed 10 million users. OpenAI, Anthropic, Google and SpaceXAI also offer agents. Khosrowshahi counts about eight relevant assistants, perhaps more than ten over time, and says all of them are talking to Uber.

**Our take.** The thesis is simple: an agent doesn't win by being able to do a task, but by doing it clearly better than the direct route. Hailing a car or reordering a dinner are exactly the cases where apps have spent a decade optimizing away friction. There, the agent competes against a button, not a twenty-step form.

That suggests where useful agents should show up first: in long, ambiguous, cross-platform tasks, such as planning a trip with flights, hotels and meetings, or handling paperwork. In those, the time savings are obvious. For what is already fast, the value lies in marginal convenience, and that doesn't change habits.

There is also a nuance worth keeping in mind. It is one week, one company, and the data is selected by Uber itself, which has an interest in presenting agents as an additional source of demand rather than a threat that could disintermediate it. Nor does it allow predictions about the future: it says nothing about what will happen when more assistants are integrated or preferences shift.

For the industry, the data cools the narrative that agents are already displacing apps, but it also leaves a strategic question open. If agents end up being the front door, platforms like Uber would go from being the destination to being just another provider behind an assistant. That is why they talk to everyone: not for today's demand, but so as not to be left out of tomorrow's intermediation layer. As industry context, it is the same pattern we see in identity and permissions: the value of an enterprise agent is decided by who acts, with what access and with what guarantees, not by the demo.

In the short term, expectations need tempering: real agent adoption is slow and depends on tangible usefulness. In the long term, the underlying logic doesn't change. When agents truly take care of the tedious stuff, they will free up human time, and that is a small but real piece of the abundance we expect. The mistake would be to confuse buzz with actual use, which is precisely what Uber's figures warn against.