Chat Is the New Booking Engine

This essay is the written form of my keynote at Future of Aviation 2026 in Helsinki. Views are my own.

“I need a flight to Konya for my mother-in-law on Saturday, after 9am.”

For twenty years, turning that sentence into a booking was the passenger’s job. Open the website. Select origin and destination. Pick dates from a calendar. Filter through dozens of options. Compare, then book. Six steps, all performed by the human, on an interface we built to make the work bearable.

Now it’s one sentence, and the AI does the rest.

The behavior already shifted

This isn’t a prediction. AI-referred traffic to e-commerce grew roughly seven-fold between January 2025 and early 2026 (Adobe Digital Insights / Shopify). People aren’t learning to use our search boxes better — they’re skipping them and asking an assistant instead.

Here’s the uncomfortable part for my industry: of the AI referral links in travel, online travel agencies get about 74%. Airlines get 26%. The last time airlines let an intermediary own the customer interface, it took decades and billions in distribution fees to claw it back. Airlines must own their AI presence — or aggregators will intermediate it again, this time at the speed of a model update.

One protocol changed the calculus

The Model Context Protocol is USB-C for AI agents: one server, any AI platform. Build once, run everywhere. Anthropic open-sourced it in late 2024; OpenAI adopted it in early 2025; by the end of 2025 it was donated to the Linux Foundation. In May 2025, Turkish Airlines launched the first airline MCP server.

That ordering matters. When a standard is forming, the early mover doesn’t just get press — it gets to make its lessons the industry’s defaults. Today our one MCP backend powers Turkish Airlines inside both Claude and ChatGPT: flight search, live status, booking lookups, Miles&Smiles services. The model holds the conversation’s context, so the Konya sentence this essay opened with simply works — no form fields, no filters.

From pages to fragments

The deeper shift is in what “user interface” means. For two decades, software meant pages: navigation, menus, scaffolding — and the user doing the work of extracting what they need. Chat-native design inverts that. You don’t ship your whole website into the conversation; you ship the one fragment the moment needs — an interactive flight card, a seat map, a price calendar — rendered right in the thread.

This is now platform doctrine, not just my opinion: the assistant platforms’ own developer guidance tells builders the same thing — don’t port your website into the chat; offer it the capabilities, one moment at a time.

What this means if you run distribution

Three things I’d tell any airline — or any company whose product is bookable:

1. Your AI presence is a channel you either own or lease. The 74/26 split is what leasing looks like.
2. The unit of product is no longer the page — it’s the capability. Inventory your APIs and ask which ones deserve to be conversational.
3. Standards moments reward the early and the careful. Move now, but treat governance as part of the product: our AI integrations authenticate with OAuth 2.1, touch only vetted APIs, and are read-only wherever possible.

Quietly, the booking engine is moving from forms to conversations. The airlines that notice early get to decide what that feels like for their passengers. The rest will find out what it feels like when someone else decides.