This is the second of three dispatches from Dublin. The first, on where airline revenue has gone, sets up the question this one answers.
For a decade the industry fought about channels. Direct or indirect, app or screen scraper, who owns the booking screen. In Dublin that fight sounded settled, and beside the point. The question now is not which channel reaches the passenger. It is whether the passenger is still the one choosing.
The decision of what to buy, and from whom, is moving to software that acts on the traveller's behalf. For an airline that folds merchandising, service, and now legal liability into a single problem. Own the agent, and you keep the customer. Miss it, and you become a commodity it picks over.
The shopper is a machine
881,776. That is how many fare queries a single AI agent fired at Etihad, the national airline of the United Arab Emirates, in one day, Shashank Nigam told the room, and not one of them ended in a booking. No traveller shops like that. None can. The thing combing the airline's fares was not a person with twelve browser tabs. It was a program, working at a scale and speed that makes the human shopper look like a relic.
Nigam, founder and chief executive of the aviation strategy consultancy SimpliFlying, was not describing a far edge of the future. When he polled the room, the overwhelming majority had already used AI to search for travel themselves. And the pressure underneath the shift is money: labour now runs at a quarter to nearly half of a US airline's total revenue, which is exactly the cost an always-on machine is built to absorb.
Who they are. SimpliFlying is an aviation strategy and marketing consultancy founded in 2008, with a team spread across several continents, that has advised more than a hundred airlines and airports on brand, customer experience and innovation. Its founder Shashank Nigam is one of the more visible voices on where the industry is heading, which is the seat he was speaking from in Dublin.
The number was not really a story about Etihad. It was a warning most of the room had not fully absorbed. Airline fare systems are already being consumed by machines they were never designed to serve, and few carriers can even tell a human shopper from a bot hammering the same endpoint. That matters because the agent does the filtering now. If an airline cannot be read cleanly and answered fast, it falls off the shortlist the machine assembles, and it falls off before any human is in the room to be persuaded. The passenger comparing tabs is yesterday's customer. Tomorrow's never sees the tabs. A machine sees them, and hands back one answer.
When the machine is wrong, the airline pays
Jake Moffatt's grandmother had just died when he asked the website chatbot of Air Canada, the country's largest airline, how its bereavement fares worked. The bot told him he could book now and claim the discount within ninety days. That was wrong. When Air Canada refused to honour it, the airline argued before a Canadian tribunal that its chatbot was a separate entity, responsible for its own words. It lost.
In Moffatt v. Air Canada, decided in February 2024, the tribunal ordered the airline to pay, and dismissed the separate-entity defence in language every carrier should read twice. A company is responsible for all of the information on its website, the tribunal held, whether it comes from a static page or a talking bot.
The case gets cited everywhere as a cautionary tale, which slightly misses its weight. It is not a story about one bad chatbot answer. It sets the floor for the whole agent era. The same reasoning that made Air Canada liable for its bot will make every airline liable for the agent it deploys, and there is no vendor contract that quietly moves that risk off the airline's own books. The sum here was small, a little over 800 Canadian dollars. The principle is not. The moment you let a machine speak for you, its mistakes become yours. Automate the answer, and you automate the accountability with it.
The fight over whose agent it is
In December, Virgin Atlantic, the British long-haul airline, switched on an assistant it named Concierge, built with the artificial-intelligence company OpenAI. By April it had become the first airline to place its app inside ChatGPT itself. Read that second move twice. The airline did not wait for the agent to arrive at its booking page. It walked into the place the agent already lived.
Who they are. Virgin Atlantic is a British long-haul airline founded in 1984, flying mainly from London Heathrow and Manchester. Delta Air Lines holds a 49 percent stake and the two run their transatlantic flying as a joint operation, and Virgin joined the SkyTeam alliance in 2023. It is a small carrier by fleet, which is part of why it can move on technology faster than the giants.
Juha Järvinen, Virgin Atlantic's Chief Customer Officer, made the bet without softening its risk. The machine, he argued, gives the airline more control and more precision than it has ever had. Then the counterweight, in his own words: only people and human-designed experiences can make you feel something, which no AI can copy.
Put the two moves side by side and the strategy reads clearly between the lines. Walking into ChatGPT is not really a distribution decision. It is a defence against being erased. If travellers start their journeys inside a general assistant, an airline that is not present there does not lose a sale, it loses the relationship that sits behind every future sale.
And Jarvinen's line about feeling is not sentiment dressed up as strategy. It is the moat. Automate the transaction, because that contest is already lost, and guard the one thing the machine cannot copy, because it is the only thing left to own. If the agent that does the choosing belongs to a platform sitting between airline and passenger, the airline ends up renting its own customer back, one query at a time.
The channel war is over. The agent war, the fight over who owns the layer that chooses, is the one that decides who keeps the passenger.
Where the agent cannot follow
The doors close, the phones drop to airplane mode, and for the next several hours the agent that did all the choosing is gone. The passenger is alone with the seat, the screen in front of her, and whatever the airline decides to put on it. For the first time in the whole journey, a brand reaches the actual human, not the software shopping on the human's behalf.
That is the inversion hiding inside the agent era, and it is the easy one to miss. Everything in these two dispatches pushes the booking, the comparison and the upsell into a layer the airline does not control. The cabin is the exception. It is the one stretch of the trip where attention is undivided, the environment is the airline's own, and no algorithm sits in between deciding what the traveller sees. Scarcity is what advertising pays for, and unmediated human attention is fast becoming the scarcest thing in travel.
The carriers have already worked this out. Delta now gives its loyalty members free onboard wifi, presented by T-Mobile and fitted on more than a thousand aircraft, and uses the sign-in to open a personalized hub of curated content and partner offers it calls Delta Sync. The wifi is the loss leader. The attention is the product, and the seatback screen that greets a passenger by name is an advertising surface no agent can stand in front of.
This is the other half of Jarvinen's point. He said only people and human-designed experiences can make you feel something the machine cannot copy. The cabin is where that happens, and it is also where the last direct line to the passenger runs. As the agent swallows the booking screen, the seat becomes the most valuable space the airline owns, not in spite of the automation but because of it.
What this leaves on the table
The passenger over the Baltic will get her coffee. The family by the dark gate will be rebooked. What Dublin settled is that a machine will increasingly be the one arranging both. The open question is whose machine. Carrying whose logo, booking whose margin, and answering, when it gets something wrong, to whom.
The advantage now sits with whoever is legible to that machine. The carriers that spent twenty years bolting intelligence onto systems built to resist it are not the ones best placed. The ones best placed are those building clean, their fares and services and answers readable by the agents that will shop them long before a traveller looks up. A system with nothing to unlearn moves faster than one defending its own history.
The skies ahead will not simply be busier. They will be quicker to judge, harder to fool, and far less forgiving of anyone who arrives legible only to humans.
The Dublin series. Three dispatches from APEX FTE EMEA and Ancillary & Retailing 2026 in Dublin. Each stands on its own.
Part 1. The Money Has Left the Ticket. Where airline revenue is moving off the ticket to the edges of the trip.
Part 2. The Intelligence Has Moved to the Agent. How AI agents now shop for the passenger, and who ends up owning the customer. (this dispatch)
Part 3. The Shop the Airport Never Builds. How an airport can earn from third-party services without building anything. (will be published soon)
Aviaedge is the analytical publication of the Association of Airports of Ukraine, read by airport and airline leadership across Europe.