Uber Says Only Four Rides Were Booked by AI Agents in October
On its Q3 earnings call, Uber disclosed startlingly low AI agent usage for rides and food orders, offering a rare, numbers-based reality check on agentic commerce hype that founders building on this promise should heed.
Independent UpShaqo analysis built from fresh, attributed sources. We explain the impact instead of repeating the announcement.
Read for leverage: focus on the workflow change, the customer problem, and the next action—not only the product announcement.
During the busiest week of Uber's third quarter, when 320 million trips and food orders flowed through its platforms, exactly four rides were booked by an AI agent. Not four thousand. Four. That number, disclosed almost in passing on Uber's earnings call, is the kind of data point that should stop founders mid-pitch-deck.
Uber's chief financial officer, Balaji Krishnamurthy, told analysts that during the week of Halloween, the company's busiest stretch, AI agents placed only 500 Uber Eats orders against that 320-million backdrop, and ride-hailing bookings through agents numbered just four. CEO Dara Khosrowshahi didn't dress it up. "It is very, very early and very limited traction that we are seeing so far," he said, adding that Uber still "hope[s] that there will be more traction."
The Gap Between Hype and Receipts
The timing makes the disclosure sting more than it otherwise might. Over the past year, the market has been flooded with personal AI agent startups promising to handle errands, bookings, and purchases autonomously. Instinct, one of the category's highest-profile entrants, raised its most recent round at a $10 billion valuation. Meta's Muse assistant became the top app in Apple's U.S. App Store, crossing 10 million users, with Meta reporting that many of those users were shopping and transacting through it. OpenAI, Anthropic, Google, and xAI have all shipped their own agent products into this same consumer narrative.
Uber's numbers are a quiet counterweight to that narrative, and it isn't alone. DoorDash told investors last week that it had received only a few hundred orders from AI agents, with the overwhelming majority of customers still placing orders directly through its app, according to the same TechCrunch report. Two of the largest on-demand platforms in the country are converging on the same conclusion: the agentic web, as currently built, is barely moving the needle on real transactions.
Why Speed Beats Delegation
Khosrowshahi offered a plausible explanation, and it's a sharp one for anyone designing agent-first products. The average time to book a ride through Uber's own app is 50 seconds. Re-ordering a favorite meal takes about 90. An AI agent has to authenticate, interpret intent, negotiate the handoff with Uber's systems, and confirm the transaction, often with extra verification steps, before it delivers any result at all. For a task that already takes under a minute, delegating to an agent doesn't save time. It adds a layer.
This is the uncomfortable truth sitting underneath the agentic commerce pitch: automation only wins when the task it replaces is slow, repetitive, or cognitively taxing. Booking a ride you've booked a hundred times before fails all three tests. The friction isn't in the task; it's in building trust and speed into a new interface that has to outperform a habit, not just a capability.
A Realistic Scenario: The Corporate Travel Platform That Bet Early
Consider a mid-sized B2B travel-management startup that spent the past year integrating Uber's and DoorDash's agent APIs into its platform, pitching enterprise clients on "zero-touch" employee travel and meal expensing. The sales pitch was compelling: an AI concierge books ground transport and catering automatically based on calendar events, no app-switching required.
Nine months in, usage data mirrors Uber's own disclosure. A handful of early-adopter employees try the agent feature once, find it no faster than opening the app they already have installed, and quietly revert to manual booking. The startup's roadmap, built around agent-initiated transactions as the core value proposition, now needs a hard pivot. The founders have two paths: narrow the agent's job to tasks where delegation genuinely saves time and cognitive load — say, batch-booking recurring team dinners or coordinating multi-city travel across time zones — or reposition the product around something agents are actually good at, like policy enforcement and expense reconciliation, with booking remaining a human-initiated action.
This is speculative, built to illustrate the mechanics rather than report an actual company, but it reflects the structural problem Uber's disclosure exposes: agent adoption lags not because the technology fails, but because the underlying task was never the bottleneck.
Why Uber Stays Bullish Anyway
Despite the numbers, Khosrowshahi remains publicly optimistic, and his reasoning is worth taking seriously even if the current usage doesn't justify it yet. He called agents "another opportunity to reach our platform" and said he expects it to be "a net positive in terms of demand on the ecosystem," according to the earnings call coverage. He noted roughly eight major AI assistants currently compete for consumer attention, a number he expects to exceed ten, and said "every single AI system company is talking to us."
That framing matters for how Uber is positioning itself, even if usage doesn't justify it yet. For a platform business, being the default fulfillment layer underneath every major AI assistant is a low-cost, high-optionality bet. Uber doesn't need agent-driven orders to be significant today; it needs to be integrated everywhere so that when and if agent usage does scale, it's the default execution layer rather than a bystander.
What This Means If You're Building on Agents
For founders and operators evaluating whether to build agent-first products or integrate agent checkout into existing platforms, Uber's disclosure offers a useful diagnostic rather than a verdict:
- Measure before you build. Instrument actual agent usage in pilot programs before committing roadmap resources to agent-first flows. Uber and DoorDash had the data; most startups haven't collected it yet.
- Target friction, not habit. Agents win on tasks involving multiple steps, comparison shopping, or coordination across services — not on single-tap actions users already perform in under a minute.
- Treat agent integration as distribution insurance, not a core product bet. Uber's own posture, show up everywhere, let demand prove itself, is a more defensible strategy for most companies than assuming agent transactions will scale on a predictable timeline.
- Watch the assistant consolidation. With Khosrowshahi projecting more than ten major AI assistants, the integration burden on merchants and platforms will keep rising even as per-assistant transaction volume stays thin, at least in the near term.
The agentic web may well arrive. But Uber's earnings call is a reminder that it hasn't yet, and that building for it prematurely carries real opportunity cost.
Sources
- TechCrunch AI — "Almost no one uses AI agents to book their Uber, company says"