Uber reveals only 4 people used AI agents to book rides in peak week

Silicon Valley has spent the past year hyping up a future where apps disappear. The pitch is simple: you whisper a quick prompt to your personal AI agent, and it books your flights, reorders your groceries, and hails your cab.

Then real-world earnings season arrived and delivered a brutal reality check.

4 rides out of 320 million

During Uber's third-quarter earnings call on Wednesday, the ride-hailing giant shared some painfully awkward data.

Chief Financial Officer Balaji Krishnamurthy told analysts that during the company's busiest week in October—Halloween week—Uber handled roughly 320 million trips and delivery orders across Uber and Uber Eats.

Out of that massive sea of rides, exactly four were booked through an AI agent.

Food delivery via Uber Eats saw slightly more activity, logging around 500 AI-initiated orders. But against the backdrop of hundreds of millions of transactions, that is essentially a rounding error.

"It is very, very early and very limited traction that we are seeing so far," Uber CEO Dara Khosrowshahi admitted on the call. "Listen, we hope that there will be more traction."

Uber is not alone. Delivery rival DoorDash reported nearly identical findings on its own earnings call last week, revealing that it had only processed a few hundred orders from AI agents so far.

The 50-second problem

These tiny numbers stand in stark contrast to the billions of dollars pouring into agentic AI.

Personal assistant startups are commanding sky-high valuations, with buzzy newcomer Instinct fetching a $10 billion price tag just a year after launch. Meta's Muse recently shot to the top of Apple's U.S. App Store with 10 million users, touting shopping and commerce capabilities. Meanwhile, OpenAI, Anthropic, Google, and SpaceXAI (Grok) are all pushing their own automated assistants into everyday workflows.

So why is nobody letting them call a cab?

Khosrowshahi pointed out the obvious flaw in the premise: calling an Uber in the app already takes an average of just 50 seconds. Reordering your favorite takeout on Uber Eats takes about 90 seconds.

In other words, AI agents simply aren't saving anyone time. In fact, prompting an agent, verifying its choices, and hoping it doesn't pick the wrong pickup pin often takes longer than just tapping your own screen.

Still, Khosrowshahi insisted he remains bullish on the category, noting there are currently around eight major AI assistants on the market and he expects that number to climb past 10. "Every single AI system company is talking to us," he added, pitching agents as just another pipe feeding demand into Uber's network.

Why it matters

Uber and DoorDash just provided the first hard, commercial sanity check on the agentic web, and the numbers speak for themselves.

We think the core problem facing consumer AI agents right now isn't capability, but friction. For lightweight, habitual tasks like hailing a ride or ordering lunch, modern smartphone interfaces have already been optimized down to seconds. Introducing an AI middleman adds latency, cognitive load, and the lingering paranoia that the model might make a rookie mistake on your dime. Unless an agent saves meaningful time on a genuinely painful workflow, everyday users have zero incentive to change their habits.

For the tech industry, this gap between venture hype and consumer behavior is dangerous. Silicon Valley is pricing agent startups at tens of billions of dollars on the assumption that agents will rapidly cannibalize the mobile app economy. But real-world transaction logs show that the consumer moat around dedicated apps remains rock solid.

AI companies clearly want Uber integration as proof that their bots can interact with the physical world. But as long as that traffic rounds down to four rides a week, the agentic web looks less like an imminent paradigm shift and more like a parlor trick in search of a problem.