Uber's Earnings Call Reveals How Thin Agentic Commerce Still Is
Uber's own Q3 numbers show AI agents booking almost no rides or food orders, forcing a harder question for operators: build agent integrations now, or wait for actual demand to materialize?
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.
Every operator weighing whether to build an "agent-ready" storefront, API, or checkout flow just got a real data point instead of a vendor pitch. On its third-quarter earnings call, Uber disclosed that during the week of Halloween — its single busiest week of the quarter, spanning roughly 320 million trips and orders — only four rides were booked by an AI agent acting on a customer's behalf, according to TechCrunch's reporting on the call. Uber Eats fared slightly better, with 500 agent-placed orders in that same week. Slightly better is still a rounding error against 320 million transactions.
The decision in front of leaders isn't whether agentic commerce will eventually matter — most product roadmaps already assume it will. The decision is timing and resourcing: how much engineering effort, partnership negotiation, and UX redesign should go toward serving autonomous AI agents today, versus waiting for adoption curves that, per Uber's own numbers, have barely left the ground.
What Uber Actually Disclosed
The facts here are unusually clean because they came straight from a public company's CFO on an earnings call, not from a startup's marketing deck. Uber CFO Balaji Krishnamurthy told analysts that during the Halloween week, four rides and 500 Uber Eats orders were initiated by AI agents, out of roughly 320 million total trips and orders. CEO Dara Khosrowshahi called the traction "very, very early and very limited," while still saying the company hopes volume grows. Those are the only hard numbers in this story, and they came from the company with arguably the most to gain from hyping agentic adoption if the trend were real.
Khosrowshahi also offered a non-trivial technical explanation for the gap: booking a ride in Uber's own app takes about 50 seconds, and reordering a favorite meal takes about 90 seconds. An AI agent navigating the same task through natural-language intent, authentication, and confirmation steps often takes longer, not shorter. That's a genuine friction problem, not just a trust or awareness problem — and it's evidence, not speculation, because it's a comparison Uber's own leadership chose to make on the call.
The Comparable Case: DoorDash's Numbers Tell the Same Story
TechCrunch's reporting notes that DoorDash, on its own earnings call the prior week, said it had received only a few hundred AI-agent-placed orders in total, with the overwhelming majority of customers still using the DoorDash app directly. Two large consumer platforms, operating in adjacent categories, independently reported the same pattern within days of each other. That convergence matters for risk assessment: it's harder to dismiss as one company's product failure when a direct competitor in a different vertical sees the identical gap between agent hype and agent usage.
For operators building on either platform's API, or evaluating whether to stand up their own agent-facing endpoints, this is a useful calibration point. The infrastructure for agentic transactions evidently exists — Uber and DoorDash both support it — but the demand side hasn't shown up yet at any meaningful scale.
Separating the Evidence From the Narrative
This is where the story gets interesting for anyone making budget decisions, because the surrounding commentary is doing a lot more work than the data supports.
What's evidence: Uber's disclosed transaction counts (4 rides, 500 Eats orders against 320 million total), DoorDash's similarly small reported volume, and Khosrowshahi's own comparison of agent task time versus app task time.
What's speculation or promotion: Khosrowshahi's forecast that the number of major AI assistants will grow from roughly eight today to more than ten, and his claim that "every single AI system company is talking to us." Those are forward-looking statements from an executive with incentive to frame agentic commerce as an emerging opportunity rather than a stalled one. Similarly, the reported growth of Meta's Muse app — which the company says topped 10 million users and became the No. 1 U.S. App Store app, with many using it to "shop and transact with businesses" — describes app downloads and self-reported usage claims, not verified transaction volume comparable to Uber's disclosure. Meta's own rollout of real-time voice and expressive avatars for Muse shows real product investment in making agent interaction feel natural, but it says nothing about whether those conversations convert into bookings at any meaningful rate.
The gap between "users are talking to an AI agent about shopping" and "an AI agent completed four ride bookings in a week" is exactly the gap operators need to interrogate before committing engineering time to agent-first flows.
Why the Gap Exists: A Short Framework
Three plausible explanations coexist, and Uber's call offers partial support for the first two without resolving which dominates:
- Speed friction. When the native app is faster than an agent-mediated flow, users have no incentive to route through an agent, regardless of how capable that agent is.
- Trust friction. Handing payment and location data to a third-party agent to book a ride involves a different risk calculus than typing into an app you already trust, and most consumers haven't made that leap yet.
- Discovery friction. Agents may simply not be surfaced as a ride-booking option in most consumer AI assistants yet, independent of whether users would use the feature if prompted.
UpShaqo's analysis: the speed-friction explanation is the most falsifiable and the most actionable for operators, because it points to a design problem rather than a behavioral or trust problem. If agent-mediated booking consistently takes longer than app-native booking, no amount of marketing or trust-building fixes the underlying math.
Practical Controls Before Committing Resources
For leaders deciding what to build or fund next quarter, a few concrete checks are more useful than reacting to headline adoption numbers:
- Benchmark agent-task completion time against your own app's fastest path before investing in an agent API — if the agent is slower, fix that first.
- Treat any vendor's "millions of users" claim about agentic shopping as a top-of-funnel metric, not a conversion metric, until transaction counts are disclosed.
- Watch the next two earnings cycles from Uber and DoorDash rather than any single quarter; one data point establishes a baseline, not a trend.
- If you operate a transactional platform, instrument agent-originated traffic now even if volume is negligible, so you have your own comparable data when volume does move.
What Remains Unresolved
Uber's disclosure answers a narrow question — how much agent-driven volume exists today — but leaves open whether that volume is suppressed by fixable friction or by a more structural reluctance to delegate purchasing decisions to software. It's also unclear whether Meta's Muse usage figures, or similar claims from other assistant makers, will ever be broken down into verified transaction counts the way Uber and DoorDash have done. Until more public companies disclose agent-specific numbers on earnings calls, operators are left triangulating between a handful of data points and a much larger volume of forward-looking statements from the companies selling the agents.