Monday morning, a last-minute trip change nearly sent me straight to AI checkout
At 9 a.m. last Monday, I was rearranging a client’s business trip from Shanghai after a flight was delayed. My first instinct was to ask an AI assistant to compare fares and book the replacement. Then I realized this was no longer just a quick price check; it might cross an airline’s restrictions on bookings made by AI agents. I got nervous too: could a few minutes saved by automation turn into client complaints and refund headaches?
This experience reminded me that AI can help, but it should not cross the point where a person needs to confirm. I also learned that making it book the ticket directly would have been a costly mistake.
This is not a sudden AI ban; it is a clearer boundary
United currently prohibits AI agents from booking tickets on behalf of travelers. In other words, AI can organize flight options, compare schedules, and draft a proposal, but the final flight choice, order form, payment, and responsibility for changes usually still need to be handled by the traveler. For independent business owners, the boundary is practical: AI can work as a “travel assistant,” but it should not be presented as a “booking agent” making decisions for clients.
I know Lin Yue, a freelance visa consultant who runs a small studio in Guangzhou. She uses AI to collect clients’ preferred departure times, airports, and budgets, then sends each itinerary to WhatsApp so the client can make the payment herself. She later told me: “I made this mistake before too. I mixed up ‘suggested flights’ and ‘ticketed and confirmed.’ When a client changed the request at the last minute, explaining responsibility was exhausting.”
What it would cost me to replicate this today
Money: An off-the-shelf AI assistant has free or low-cost plans for roughly RMB 0–100 per month; more complex collaboration setups cost extra. Time: I would spend 30 minutes defining the boundaries, then another 1–2 hours writing “AI recommends only; a human must confirm” into the workflow. Technical barrier: No programming required, but I need to distinguish a recommendation from a final order and check refund, baggage, and authorization rules. First step: Open the AI assistant I already use and create a “Client Itinerary Suggestions” template that only outputs candidate flights, prices, and risk notes, without connecting it to payment.
I would start with a spreadsheet and email workflow rather than build an automated booking system. I would also keep a human confirmation record covering client authorization, the final selection, price, and payer. If something goes wrong, I can see exactly which step needs to be redone. Platform policies change, so I would check the airline’s latest instructions before departure.
How I would set this up at each stage
If I were just starting out, I would let AI organize requirements and candidate options while handling every booking action myself. There is no need to try this immediately; manually checking two or three flights can be just as reliable.
If I already had one or two stable clients, I would build a semi-automated workflow: AI generates the proposal, the consultant sends a confirmation link, and the client completes payment and seat selection. I would also spot-check the rules regularly.
If I were scaling up, I might connect an official airline API or licensed partner, with permissions, audit logs, and a refund process clearly documented. This tool is not essential for everyone, and I would not add system overhead just to save a few minutes.