A Test, Not a Failure
During one of its busiest weeks in late October, with around 320 million trips across its ride-hailing and delivery platforms, Uber reported a surprisingly small number of transactions initiated by third-party AI agents. Out of this massive volume, only
about 500 Uber Eats orders and a mere four rides were booked via these emerging AI assistants. DoorDash reported similarly low numbers, with only 'several hundred orders' placed by agents over a much longer period. Far from being a sign of AI's shortcomings, these numbers reflect a deliberate 'crawl, walk, run' strategy. For tech giants, launching new technology isn't about making a big splash; it's about managing risk, gathering data, and ensuring that any automated system is safe, reliable, and doesn't damage the user experience or brand trust.
What is an AI Agent, Anyway?
It's important to distinguish AI agents from the simpler chatbots we've grown accustomed to. A chatbot can answer questions or look up information. An AI agent, however, is designed to take action on your behalf. It can understand a complex command like, 'Order my usual Friday night pizza and a salad, and have it delivered around 7 PM', and then execute the entire process—from opening the app and selecting items to completing the payment. This ability to act autonomously within other applications is what makes agents so powerful, but also so risky. A flawed chatbot gives a bad answer; a flawed agent could order the wrong thing, charge the wrong card, or create a customer service nightmare. This is why most companies are moving slowly, with nearly 90% delaying AI rollouts due to data security and governance concerns.
Uber’s Game of Cautious Control
Uber's CEO, Dara Khosrowshahi, has a unique perspective shaped by his past experience at Expedia, where dependence on Google for traffic was a constant challenge. He has explicitly stated, 'I don't want to go back there'. This sentiment drives Uber's strategy. While the company is in discussions with all major AI agent developers, it is carefully negotiating the 'terms of trade'. Uber wants to ensure it controls the customer experience and doesn't become a commoditized service layer for another company's AI. By slowly integrating and building its own AI tools, like an in-app agent that has more context about a user's travel plans, Uber aims to keep customers within its own ecosystem. Khosrowshahi also noted that ordering an Uber is already fast—about 50 seconds—so an agent needs to provide significant added value to be adopted.
DoorDash and the Discovery Engine
DoorDash is taking a multi-faceted approach. While its external agent numbers are low, it has been aggressively developing its own AI features. The company launched 'Ask DoorDash', an in-app natural language search tool to help users find what they want without endless scrolling. It has also experimented with a text-to-order feature and integrations with platforms like ChatGPT to help with meal planning. More significantly, DoorDash has successfully deployed a generative AI solution to handle hundreds of thousands of daily support calls from its 'Dasher' delivery drivers. This frees up human agents for more complex issues and shows where AI is having a real, immediate impact: improving internal operations and support systems, rather than radically changing the consumer-facing ordering process overnight.
Airbnb’s AI Concierge Dream
Unlike the quick, transactional nature of food delivery, Airbnb sees AI playing a very different role. For them, the goal is to create an 'AI concierge' that helps users plan entire trips. Booking travel is a higher-consideration process where users spend more time researching. An AI agent can shine here by learning a user's preferences to offer personalized recommendations for stays and experiences. Airbnb has already seen significant success using AI in customer support, where its assistant resolves a large percentage of inquiries without human intervention, leading to faster resolutions and lower costs. This strategy proves that the best use of AI isn't a one-size-fits-all solution; it’s about applying the technology to solve a specific business problem, whether that's operational efficiency or enhanced trip planning.
















