Uber AI Agents Experiment: Why Almost Nobody Uses AI to Book Rides
TL;DR
- Uber says just four people used an AI agent to book rides during its busiest week last quarter.
- The experiment highlights consumers’ reluctance to let AI handle payments, transportation choices, and other real-world tasks.
- Trust, reliability, accountability, and the convenience of Uber’s existing app may be bigger obstacles than the technology itself.
A striking figure from Uber’s latest results has put a spotlight on the gap between the promise of AI agents and their everyday use: only four people booked rides through an AI agent during the company’s busiest week last quarter.
The number is tiny compared with Uber’s overall scale. The company serves millions of riders and processes a vast number of trips each day, yet almost nobody chose to delegate a simple ride booking to an artificial intelligence assistant.
The result does not necessarily mean Uber’s AI tools failed technically. Instead, it offers an early look at a more difficult challenge for the technology industry: persuading people to trust AI with actions that affect their money, schedules, safety, and physical movements.
A Small Number With a Larger Message
AI agents are designed to do more than answer questions or generate text. They can potentially search for services, make decisions based on a user’s instructions, complete transactions, and interact with several apps on the user’s behalf.
Booking a ride is a straightforward example. A user might ask an assistant to arrange transportation to an airport, select an appropriate vehicle, and pay for the trip without opening the Uber app.
In practice, however, consumers have shown limited interest in handing over that responsibility. Uber’s figure suggests that the company’s AI-agent experiment remains at a very early stage, even as technology firms increasingly present autonomous digital assistants as the next major computing platform.
The challenge is especially visible in transportation. A ride booking is not merely a search query. It involves a pickup location, destination, timing, price, driver availability, payment details, and potentially sensitive location information. A mistake can result in a missed appointment, an unexpected charge, or a rider being sent to the wrong place.
Convenience May Not Be Enough
One reason for the low adoption may be that Uber’s existing app is already relatively easy to use. Opening the app, entering a destination, checking the price, and confirming a ride can take only a few seconds.
That leaves AI agents with a difficult value proposition. If an assistant saves only a small amount of time while introducing uncertainty about what it will do, many users may prefer to remain in control.
Consumers are also accustomed to seeing Uber’s interface before confirming a trip. They can review the vehicle type, estimated arrival time, fare, pickup point, and destination. An AI agent may be able to provide the same information, but users may feel less comfortable when the transaction happens through a conversational interface or in the background.
The more familiar process can therefore be more attractive than the theoretically more advanced one.
Trust Is the Core Barrier
The biggest obstacle for AI agents is likely to be trust rather than awareness.
People may be comfortable asking an AI assistant to recommend a restaurant or summarize an article. They may be much less comfortable allowing the same system to spend money or make a decision with real-world consequences.
That distinction is important. A wrong answer in a casual conversation can be ignored. A wrong ride booking can create immediate practical problems.
Users may worry that an agent will:
- Choose the wrong pickup location
- Misinterpret a destination
- Select an unnecessarily expensive ride
- Book at the wrong time
- Apply the wrong payment method
- Make duplicate reservations
- Fail to explain a surge price
- Share location or financial information more broadly than expected
Even when an agent performs correctly most of the time, users may focus on the consequences of the occasional failure. Reliability requirements are much higher when software is authorized to act instead of merely provide information.
Who Is Responsible When AI Makes a Mistake?
Accountability presents another unresolved issue.
If a user books a ride directly through Uber and something goes wrong, the company’s support process is clear. If a third-party AI agent makes the booking, responsibility could become more complicated. The agent provider, Uber, the payment platform, and the user may all be involved.
A customer might ask: Who is responsible if the agent confirms the wrong address? Who refunds a ride booked twice? What happens if the system misunderstands an instruction such as “get me there as cheaply as possible” and chooses a longer or less convenient option?
These questions matter even more as companies build systems that can take actions across multiple services. Consumers may hesitate until platforms clearly explain how errors are handled and how unauthorized actions can be reversed.
Payments Raise the Stakes
Payment is a particularly sensitive part of the AI-agent experience.
Many users already hesitate before entering card details into unfamiliar websites. Allowing an autonomous system to access stored payment information introduces a different level of concern. The agent must know when it has permission to spend, how much it may spend, and whether a new charge requires confirmation.
A natural-language instruction can also be ambiguous. “Book me a ride to the airport” leaves open questions about vehicle type, timing, price, and the airport’s terminal. A human using the Uber app resolves those details by viewing options. An agent must either make assumptions or repeatedly ask for clarification, reducing the convenience it is supposed to provide.
Companies may eventually address this through spending limits, approval prompts, transaction previews, and detailed activity logs. Until those safeguards become familiar, consumers may prefer to keep financial decisions in their own hands.
Privacy and Security Concerns
Ride-hailing generates sensitive data, including home and work locations, travel patterns, pickup times, and payment activity. Using an AI assistant as an intermediary may create additional questions about where that information is stored and who can access it.
Users may not know whether their instructions are retained by the AI provider, used to improve models, or shared with the service being accessed. They may also worry about account takeovers, prompt manipulation, or malicious instructions embedded in connected services.
AI agents introduce new security risks because they can potentially read information and take actions. A conventional chatbot may produce a misleading answer; an agent could potentially act on malicious or incorrect data.
For mainstream adoption, companies will need strong authentication, transparent data policies, permission controls, and ways to inspect exactly what an agent did.
The Reliability Problem
AI systems can be impressive while still being unpredictable. They may understand a straightforward request in one situation and misinterpret a similar request in another.
That unpredictability is poorly suited to services that require precision. A ride-hailing agent must reliably identify addresses, understand time zones and dates, distinguish between similarly named destinations, and handle last-minute changes.
It must also cope with real-world conditions. Prices can change rapidly. Drivers may cancel. A pickup location may be inaccessible. An airport may have different pickup zones. A user may need to update a booking while walking through a crowded station.
An agent that cannot communicate these complications clearly may create more work than it saves.
The “Four People” Figure Is Also a Product Lesson
Uber’s experiment provides a useful reminder that enabling a feature is not the same as creating demand for it.
Technology companies often assume that consumers will adopt autonomous tools once they become technically capable. But adoption depends on whether the feature solves a meaningful problem, fits existing habits, and gives users confidence that they remain in control.
In Uber’s case, the potential benefit may be too small for most people to change their behavior. A user who already has the Uber app installed may see little reason to insert another layer between themselves and the service.
AI agents may have a stronger advantage in more complicated tasks, such as coordinating travel across airlines, hotels, trains, and ride services. They could also help people who have accessibility needs or who regularly manage transportation for family members.
But those use cases will require more than a conversational interface. They will require dependable execution and clear controls.
Why Businesses Still Want AI Agents
Despite the weak early consumer uptake, companies have strong reasons to keep experimenting.
AI agents could eventually increase the number of transactions completed through digital platforms, reduce customer-service costs, and make services easier to access across different applications. Instead of asking users to navigate several apps, an agent could coordinate the process from a single conversation.
For Uber, an agent might eventually help users plan an entire journey, combining rides with public transportation, flights, or deliveries. It could remember preferences, anticipate recurring trips, and provide support when plans change.
The commercial opportunity is therefore larger than simply replacing a few taps in the Uber app. Companies are competing to become the infrastructure through which AI agents discover, compare, and purchase services.
That competition explains why firms continue to invest even when initial usage numbers are modest.
The Industry’s Chicken-and-Egg Problem
AI agents face a classic adoption cycle. Consumers are reluctant to use them until they are reliable and widely supported. Businesses are reluctant to invest heavily until enough consumers use them.
For agents to become useful, they need access to many services. For services to grant that access, they need confidence that agents will represent their brands accurately, protect customer data, and avoid unauthorized transactions.
A fragmented landscape could make the problem worse. Different companies may use different standards for authentication, permissions, payments, and customer support. Users may be forced to manage separate settings for each agent and service.
Common technical and legal standards could help. So could clearly defined rules about consent, liability, refunds, and data use.
What Would Make People Try It?
AI agents are more likely to gain traction if they give users precise control instead of asking for unlimited autonomy.
Useful features could include:
- A confirmation screen showing the fare, route, vehicle, and payment method
- Spending limits and restrictions on ride types
- Temporary permissions that expire after one transaction
- A detailed record of every action taken by the agent
- One-tap cancellation and refunds
- Strong identity verification for sensitive requests
- The option to require approval whenever prices exceed a set amount
- Clear explanations when the agent makes assumptions
These controls could make AI feel less like an unpredictable assistant and more like a supervised automation tool.
The most successful systems may not be fully autonomous. They may operate in a middle ground, handling routine steps while asking users to approve important decisions.
A Slow Start, Not Necessarily a Dead End
Four AI-booked riders in a peak week is an awkward result for an industry eager to promote agentic computing. But it is too early to treat the experiment as a definitive verdict on AI-powered commerce.
Many technologies take time to move from demonstrations to habits. Consumers may need to see repeated evidence that agents are accurate, secure, and easy to correct. They may also need a stronger reason to use them than simply saving a few taps.
The early Uber results suggest that the central question is not whether AI can book a ride. It is whether people believe AI can book the right ride, charge the right account, protect their information, and fix the problem when something goes wrong.
Until the answer is obvious, most riders are likely to keep doing what they already do: open the Uber app, check the details themselves, and press the button.
Get All The Latest Updates Delivered Straight To Your Inbox For Free!