What Are AI Agents, Anyway?
First, let's clarify what we mean by an 'AI agent'. Unlike a simple chatbot that follows a script, an AI agent is a more autonomous system. It can understand natural language, maintain context, and perform tasks on your behalf across different applications.
Think of a personal assistant that can not only answer a question but also book a ride, order your dinner, or coordinate a group grocery order by interacting with other services. The ultimate promise is a world where you simply state your goal—'get me a ride to the airport' or 'order my usual Friday night pizza'—and the agent handles the rest without you needing to open an app.
The 500-Order Reality Check
During Halloween week, one of its busiest periods, Uber processed around 320 million trips and deliveries. Out of that massive number, external AI agents were responsible for a 'grand total of four trips' and 'around 500 orders' for Uber Eats. Uber's CFO, Balaji Krishnamurthy, called it 'very limited traction'. Rival DoorDash reported a similarly low figure, with CEO Tony Xu stating that third-party agents had generated only 'several hundred orders' in total since they launched. Airbnb's CEO Brian Chesky also described agent-driven bookings as 'very non-material'. On the surface, these numbers seem like a stunning flop for a technology hyped as the next big thing. But the reality is more nuanced.
The High Cost of Getting It Wrong
The slow rollout is not due to a lack of technology, but an abundance of caution. For customer-facing brands, deploying an AI that makes mistakes can be catastrophic. An AI agent that 'hallucinates'—or invents information—can cause serious brand damage. Air Canada, for example, was legally required to honor a refund policy that its support chatbot completely invented. For Uber, DoorDash, and Airbnb, whose services are woven into the fabric of daily life, the risk of an AI misquoting a price, booking the wrong service, or mishandling a customer complaint is too high. These companies understand that customer trust, once broken, is incredibly difficult to repair. A single bad AI-driven experience can go viral, leading to reputational damage that far outweighs the potential efficiency gains.
A Playbook of Cautious Experimentation
Instead of a full-scale launch, these tech leaders are following a shared playbook: test, learn, and control the environment. They are all developing their own in-house AI tools while carefully observing how third-party agents interact with their platforms. Uber, for instance, has AI co-pilots that assist human customer service agents by summarizing conversation histories and suggesting responses, improving efficiency without removing the human entirely. DoorDash launched its own 'Ask DoorDash' and text-to-order assistants, allowing it to manage the user experience directly. Meanwhile, Airbnb is building a proprietary AI travel agent, slated for 2027, designed to help users plan entire trips, arguing that general-purpose agents can't handle the visual and social nature of travel planning. This strategy allows them to gather data and refine the technology in a controlled setting.
Controlling the 'Terms of Trade'
There's another strategic layer to this caution: platform control. Uber CEO Dara Khosrowshahi has been open about not wanting to become overly dependent on third-party platforms for customers, drawing a parallel to the travel industry's reliance on Google. By developing their own AI and negotiating terms with external agent creators, these companies are positioning themselves to control the 'front door' to their services. They want to avoid a future where they become a commoditized fulfillment layer for a powerful, overarching AI agent that captures all the customer data and loyalty. The minuscule transaction numbers are a clear signal of who currently owns the customer relationship—and these platforms intend to keep it that way.
















