Leading with Agentic AI: Executive Perspectives with Travolution
Summary
This panel session titled 'Leading with Agentic AI: Executive Perspectives with Travolution' was moderated by Kate Harden-England, editor of Travolution (Europe's leading digital travel and travel technology publication), and featured James Lemon from Stripe and Stephanie Madee from Amadeus. The approximately 21-minute discussion focused on the current state and near-term trajectory of agentic commerce in travel.
James Lemon opened by framing agentic commerce as the natural evolution from natural language discovery to actual booking and purchasing — moving from 'let's discover' to 'let's book and buy.' He referenced Stripe co-founders Patrick and John Collison's 2025 annual letter, which described agentic commerce in five levels, noting that the industry is currently only at levels one or two — primarily search and description using natural language.
Stripe's key initiative is its Agentic Commerce Protocol, developed over approximately six months in collaboration with OpenAI, which Stripe subsequently open-sourced on GitHub. Lemon compared it to Linux as a new open standard — allowing any merchant to participate without being locked into Stripe as a payment processor. The protocol creates a checkout-like experience inside large language models such as ChatGPT, generating a data package of payment details to be forwarded to merchants. As of the session, Stripe had shipped connectors to both the Visa and Mastercard protocols, and had gone live with shared payment tokens with Klarna and Affirm.
Stephanie Madee of Amadeus highlighted a critical gap: current AI commerce protocols are not travel-ready. Existing protocols rely on fixed reference numbers (such as a specific TV model number), which do not exist in travel. In travel, a reference number is only created at the moment of search — for example, calling an airline to generate an offer and an offer ID. The notion of servicing, central to travel, is also absent from current protocols. Amadeus has begun assessing what changes are needed to make these protocols travel-ready. Amadeus has built a new AI-friendly API easy for both AI assistants and travel sellers to consume, and is offering MCP (Model Context Protocol) on top of several services including its leisure platform. Amadeus also acquired Skylink to strengthen agentic orchestration capabilities, and is partnering with Microsoft and Google on co-innovation.
The panel discussed where agentic AI is delivering value today. Both panelists agreed that post-booking automation is the most immediate near-term opportunity. Madee cited Amadeus's investment in voice-to-text, text-to-voice, and automated post-booking workflows. Lemon cited a large Australian travel agency (unnamed) that saved 20% of total call center time simply by enabling guests to modify trips online. Lemon also described a use case for schedule change management: an AI agent can detect a major schedule change, assess the impact, search for alternatives, contact the traveler, propose options, and automatically book — a process that previously took travel agents 20 to 40 minutes per case and often could not be billed to the client.
However, Lemon argued that the 'real opportunity' lies in booking and discovery — what he called 'a revolution in the way travel is searched, booked, and paid.' He predicted an explosion in the 'connected trip,' where travelers start with an experience, meal, or destination feeling rather than a flight and hotel, and everything else tucks in around that.
Madee pushed back on the brand-visibility strategy of getting discovered within general AI assistants like Gemini or ChatGPT, expressing skepticism: 'I don't believe that this is how it will work in future.' She argued that users are unlikely to proactively request a specific brand within a general AI assistant.
For immediate action, Madee's top recommendation was 'get your data sorted,' explaining that brands now need to convince AI systems — not just humans — and AI assistants cannot be swayed by emotions or visual cues. Lemon's recommendation was to build an AI direct booking experience in-house or with a partner, citing the AORE app launched recently as an example of a brand building the necessary 'muscles' — teams that understand how to experiment, deploy, and iterate with AI in a live customer context.
The panel concluded with the three pillars Lemon identified as necessary to win in agentic commerce: Discovery (strong brand, findability), Accuracy (live availability, rates, and inventory blended with unstructured data), and Payments (frictionless, safe transfer of money from travelers to merchants). Final one-word takeaways: Lemon said 'experimentation'; Madee said 'data.'
Key Takeaways
- •Stripe co-founders Patrick and John Collison described agentic commerce in their 2025 annual letter as having five levels, and James Lemon confirmed the industry is currently only at levels one or two — primarily natural language search and description. The journey to full delegation, where an agent books autonomously on the traveler's behalf, is still ahead, but Stripe is betting it will accelerate quickly, especially for repeatable tasks like routine business travel.
- •Current AI commerce protocols are fundamentally not travel-ready, according to Stephanie Madee of Amadeus. Protocols like those used by major US retailers rely on fixed product reference numbers, which do not exist in travel — a reference number in travel is only created at the moment of search (e.g., when calling an airline to generate an offer ID). The concept of servicing, which is central to travel, is also entirely absent from current protocols. Amadeus has begun formally assessing what changes are needed to close this gap.
- •Stripe spent most of 2025 building its Agentic Commerce Protocol in a six-month collaboration locked in a room with OpenAI. The protocol was subsequently made fully open source on GitHub, allowing any merchant or platform to contribute feedback and shape updates. Lemon compared it to Linux — 'the new standard' — meaning merchants are not required to use Stripe as their payment processor to benefit from the protocol.
- •As of the session, Stripe had shipped connectors to both the Visa and Mastercard protocols and had gone live with shared payment token integrations with Klarna and Affirm. Lemon described the core mechanism: a checkout-like experience is generated inside an LLM like ChatGPT, which packages payment data and forwards it to the merchant to charge — removing the need for each merchant to build directly to LLM-specific standards.
- •Amadeus has built a new AI-friendly API consumable by both AI assistants and traditional travel sellers, and is offering MCP (Model Context Protocol) access on top of several services including its leisure platform. Amadeus acquired Skylink to strengthen agentic orchestration capabilities and is co-innovating with Microsoft and Google. CEO Louise Moroto stated in Q3 2025 earnings that AI assistants are 'unlikely to become merchant or full content aggregators on their own,' framing this as expanding rather than eroding Amadeus's role.
Transcript
And ladies and gentlemen, it's a food for thoughts now leading uh with Agentica. Executive perspective is Travolution. And who could do this better than my very dear colleague Kate Harder England, the editor of Travolution. Kate, with pleasure, this stage is yours. Thanks for being here. >> Thank you very much, Durk. Uh good afternoon, everyone. Um, as Durk said, I'm Kate Harden, England, editor of Travolution, Europe's leading Are we back in the room? Yes. Digital travel and travel technology p...
