Travel AI's consumer breakthrough will be measured by how much responsibility travelers hand over — not by how many steps an agent automates, according to a Skift analysis that reframes how sellers of travel should judge their AI investments.

The piece draws on a distinction made by Ben Thompson in The iPhone's Last Stand: enterprises pay for employee time, while consumers mostly want tasks and coordination burdens to move somewhere else. Applied to travel, that shifts the benchmark from productivity to convenience — work the traveler no longer has to do.

What does the responsibility ladder look like?

The analysis lays out four steps of agent responsibility:

• Answer — explain options to the traveler.
• Compare — narrow the choices.
• Execute — book, with approval.
• Assure — stay responsible after the transaction by monitoring flights, protecting loyalty benefits, chasing refunds, and repairing disruptions.

The fourth step is where the real value sits, the analysis argues, because a confirmation number does not mean the problem is finished. Airlines have already largely solved the first-order disruption: Delta may automatically rebook a traveler after a cancellation, United can surface replacement options in its app, and corporate travel systems can intervene before the traveler ever calls an agent.

The harder question is everything around it. Does the replacement flight keep the family together? Does the hotel need another night? What happens to the airport pickup? Is there a refund to chase — and will anyone notice if the replacement flight changes again?

A truly convenient agent would take responsibility for that whole chain and return only when a real tradeoff needs approval. The traveler would stop being the disruption manager — a sharper test, the analysis suggests, than how powerful the agents become.

Where does Expedia draw the line?

Convenience is selective. Expedia CEO Ariane Gorin said at Skift Global Forum that she does not expect a large share of bookings to become one continuous conversation, because planning is multi-turn, often involves other people, and sometimes benefits from slowing down. Expedia has moved away from one universal agent toward narrower agents for specific parts of planning and shopping.

That architectural retreat matters for distribution. If the booking flow fragments into task-specific agents, sellers will need to decide which tasks justify automation and which still require a human in the loop.

What should still require approval?

The analysis is explicit about the design question: routine work such as re-entering passenger details or filing a refund claim can be automated, while splitting a family across flights or giving up a loyalty benefit should still require traveler approval. Reliability, privacy, access to underlying services, and human fallback remain the binding constraints on how far that automation can go.

For platforms, airlines and intermediaries building agentic booking, the analysis implies a clear priority order: the post-transaction layer — monitoring, repair and refunds — is the underbuilt rung of the ladder, and the one where travelers would notice the difference first.