Agentic commerce was supposed to be this year’s breakthrough, the moment AI agents began quietly shopping, booking rides, and ordering food on our behalf. Instead, the first hard numbers are starting to tell a more modest story. On their Q3 2026 earnings calls in late October and early November, Uber, DoorDash, and Airbnb each disclosed how much of their business is actually coming from third party AI agents, and the figures are far smaller than the agentic commerce narrative suggested. Uber’s finance chief told investors that agents produced just four rides and about five hundred Uber Eats orders out of 320 million rides and deliveries in a single week. DoorDash reported a few hundred orders in total since launch, and Airbnb called its agent bookings a non material share of its roughly half a million daily reservations. For anyone building or investing in AI agents, these numbers matter more than another demo video. This article walks through what the earnings calls revealed, why adoption is lagging behind the hype, and what builders and business leaders should take from the gap between the promise and the current reality.

The Numbers Behind the Agentic Commerce Slowdown

The data comes directly from the companies with the most to gain from agentic commerce working. On Uber’s November 4 call, CFO Balaji Krishnamurthy said the company saw agents book only four rides and roughly five hundred Uber Eats deliveries during a single Halloween week that included 320 million total trips, a share so small it rounds to zero percent. DoorDash CEO Tony Xu told investors on October 29 that agents had generated maybe several hundred orders in total since launch, out of a business that processes tens of millions of orders. Airbnb CEO Brian Chesky, on a November 5 call, described agent driven bookings through Meta’s Muse and OpenAI’s agent products as a very non material number against roughly half a million daily bookings. All three companies pointed out that the relevant third party agents only launched in September, so the sample size is still small. Even so, the gap between the attention this idea has received and the transactions it is actually producing is striking, and it is a useful reality check for anyone sizing a market opportunity around AI agents doing commerce on a person’s behalf.

Why AI Shopping Agents Aren’t Converting Users Yet

The executives and analysts closest to the data offered a consistent set of explanations, and none of them blame the technology outright. M7 Innovations CEO Matt Maher argued that many of the tasks agents target, like booking a restaurant reservation, were already fast and simple, so agents are, in his words, solutions looking for problems. Speed reinforces that point: Uber CEO Dara Khosrowshahi noted that a typical Uber order takes about 50 seconds and an Uber Eats reorder about 90 seconds on the app, which leaves little incentive to hand the task to an agent and wait for it to act. Airbnb’s Chesky made a different argument specific to travel, saying planning a trip is aspirational and that roughly 80 percent of travelers enjoy the research itself, so a general purpose agent that skips the browsing may be solving a problem users do not want solved. Talkdesk’s Michael Klein pointed to trust as the deeper issue: many consumers are not yet comfortable letting an agent make choices and handle payment details on their behalf without full visibility into how that data is used. Put together, the lesson is that agentic commerce is not failing on capability, it is running into human habits, trust, and the fact that some tasks genuinely do not need automating.

What Builders Should Learn From the Agentic Commerce Adoption Gap

If you are building AI agents or deciding whether to adopt them, the practical takeaway is to separate genuinely painful, high friction tasks from ones that are already quick. Uber and DoorDash are not retreating from AI agents, they are investing in their own first party versions instead, such as Uber’s Cart Assistance, which fills a shopping basket from a handwritten list or a recipe screenshot, and DoorDash’s Ask DoorDash and Text DoorDash tools. Internal reports suggest these first party baskets run roughly twice the size of an average order, which is a meaningful signal: agents may add the most value inside a platform a customer already trusts, rather than as an outside layer negotiating on the customer’s behalf. We have written before about the importance of AI agent guardrails for safe enterprise automation, and this data adds another dimension to that lesson: an agent also needs a clear job that users actually want off their plate. For AI agent startups, this points toward building agents that reduce real friction, such as multi step research, comparison, or data entry, rather than automating a single tap a user is happy to make themselves. It is also a reminder to validate demand with real usage data before building a go to market plan around AI agents replacing an entire category of consumer behavior.

The Agentic Commerce Opportunity Is Still Ahead, Just Not Yet

None of the companies involved are writing off agentic commerce. Khosrowshahi said Uber is in active discussions with third party AI agent companies and described every major AI system provider as currently talking to Uber about terms. Airbnb has its own travel planning and reservation agent due to roll out early next year, built around the company’s own data and trust relationship with guests. This mirrors a pattern we explored in our piece on AI agents operating without traditional APIs: the companies with the most customer trust and data tend to shape how agentic access actually gets built, rather than opening the door wide to any outside agent on day one. The likely near term path is less about a single autonomous agent replacing an app and more about platforms opening narrow, well defined channels for agents to transact through, while keeping their own first party agents as the primary experience. For founders building in this space, the opportunity is real, but the current data suggests it will be won by solving specific friction points with clear trust and speed advantages, not by promising a general purpose agent that can book anything.

Conclusion: What the Data Means for You

Three things stand out from this quarter’s earnings calls. First, the agentic commerce numbers reported by Uber, DoorDash, and Airbnb are still extremely small relative to total transactions, even months after major agent products launched. Second, the executives closest to the data blame trust, speed, and task fit rather than the underlying AI capability. Third, the platforms themselves are responding by building their own first party agents rather than waiting for third party ones to take over. If you are building or evaluating AI agents for your own business, let real usage numbers, not demo reels, guide where you invest next. For more analysis of AI agents, automation, and the tools worth watching, explore more at BigAIAgent.tech. If adoption is still this early, what would it actually take for you to hand a booking or a purchase over to an agent completely?