Monday’s booking failure showed me where an assistant’s “usefulness” ends
Last Monday morning, I was helping a client catch a flight from Shanghai to Beijing. I gave my AI assistant the details: “Wednesday morning, a 1,200-yuan budget, and I need to be able to attend an afternoon meeting.” It quickly found three suitable flights. But when we reached the payment page, United’s system rejected the personal AI agent outright. After 20 minutes of troubleshooting, I manually completed the order. That moment made it clear: an assistant having access to my account does not mean it can handle everything for me.
The real conflict is over who owns the customer relationship
United Airlines is restricting personal AI agents from parts of its booking flow. The logic is not hard to understand: when an assistant knows my preferences and decides which airline I spend money with, the airline may lose its chance to communicate with me directly.
My friend Lin Yue runs a one-person travel advisory business in Hangzhou. She lets her assistant shortlist flights and compare change-and-refund rules, but she leaves payment to the client. She has also been stuck: when she first allowed the assistant to place orders directly, an error on the page made a client think she was unprofessional. She switched to “AI proposes, a person confirms.” Her response time did not slow much, and clients felt more reassured. Several travel-management teams now use a similar split: AI organizes the information, while a human makes the final commitment.
Try it today: 0 yuan, 30 minutes, no coding
Money: 0 yuan. Time: about 30 minutes. Technical barrier: knowing how to copy and paste requirements. First step: open the AI chat tool you already use, start a new conversation, choose “Explore” mode, then enter: “Compare the price, timing, and change/refund conditions for three flights. Do not book yet.”
If you use a travel agency or airline app, I can also have AI help organize the filters before I manually open the matching flight pages. It can generate a comparison table so I can check the total price, checked baggage, and change fees myself. This setup is not essential for everyone; waiting is perfectly reasonable. Handing sensitive accounts and personal identity documents to AI is not a good trade right now.
Match the workflow to your stage—full automation is not the goal
Just starting: If I do not have stable clients yet, I would use AI to list requirements and compare options while completing bookings myself, establishing a reliable process first.
With 1–2 clients: If I regularly arrange trips for clients, I would separate screening, comparison, reminders, and final confirmation, while keeping payment and identity information under human control.
Scaling up: If I am preparing to let AI handle more orders, I would first log every case where an airline or platform blocks it, then assess whether a formal travel-management service is worth integrating.