Travelport has spent the past year building a case that AI agents cannot safely book a flight without a deterministic system underneath them. On 1 July, that case moved from conference-stage argument to live commercial deployment.
Travelport’s Trip Services API is now integrated into Orchestra, the Transport Hub at the centre of the Travelsoft Group’s platform ecosystem. The integration puts Travelport’s airline content, both legacy EDIFACT and modern NDC, in front of more than 400 travel agencies across Europe through Orchestra and Travel Compositor, Travelsoft’s dynamic packaging platform. Travelsoft says the rollout will extend progressively to its other platforms, including airQuest, Atcore, Tigerbay, Traffics, and Travel Connection Technology. The scope has already grown past basic content distribution to cover post-sale servicing: exchanges, involuntary changes, ancillary services, ticketing, upsell, and multi-city booking.
On its own, this is a distribution story: a mid-sized GDS extending its European agency reach through a SaaS aggregator. But the timing is what makes it worth a second look for anyone tracking how AI is being wired into airline distribution.
The argument behind the integration
Travelport’s chief product and technology officer, Andrew Jordan, laid out the underlying thesis at the TravelTech Show in London days before the Orchestra deal went live. His argument is structural, not promotional: large language models predict the next probable token, they do not query live inventory, and when a model states a seat is available, that is an inference rather than a lookup. In travel, an inference that turns out wrong becomes a ticket to the wrong city or an unauthorised charge, not just a bad answer.
Travelport’s response has been to split the stack into two layers. A conversational layer, built on Anthropic’s Claude, interprets what a traveller wants and proposes options. A deterministic layer, TripServices, executes the booking against live inventory. The two are connected through the Model Context Protocol (MCP), the open standard Anthropic introduced in late 2024 and handed to the Linux Foundation’s Agentic AI Foundation in December 2025. MCP gives the reasoning layer a standard way to call the booking layer without needing to infer what the outcome should be.
Cognizant is providing the engineering capacity to migrate Travelport’s older codebase into this architecture, work that involves parsing decades of embedded fare logic as much as building new interfaces.
The Orchestra integration is the first deployment at meaningful scale. More than 400 agencies now sit on infrastructure built around that separation of reasoning from execution, whether or not each agency is actively running an AI booking layer on top of it yet.
Why the trust gap matters more than the architecture
The engineering case is only half the story, and arguably the less contested half. Building a deterministic booking engine solves the technology problem. Convincing travellers to trust it is a very different challenge.
An Expedia Group survey of more than 5,700 adults published in April found that only 8% of travellers are willing to let AI complete a booking on their behalf, even as more than half are comfortable letting AI suggest options. A separate Booking.com survey found the large majority of respondents have concerns about AI outputs, with roughly a third saying they trust them fully. Documented failure cases already exist: AI systems reporting bookings as confirmed while still pending, cancellations processed without traveller consent, and unauthorised charges to stored payment methods.
That gap between industry enthusiasm and traveller caution is the real constraint on how fast this architecture translates into revenue. Jordan’s own framing draws a useful line: at the discovery end of the funnel, an AI agent getting things right most of the time is tolerable. At the execution end, where a booking becomes a financial commitment, the margin for error is effectively zero. Distribution infrastructure that reduces AI hallucination risk addresses one side of that equation. It does not by itself close the trust gap on the traveller side.
The competitive read
Travelport is not alone in betting that the winning architecture separates conversation from execution. Sabre took a similar approach with Mindtrip’s Flights product, launched in May in partnership with PayPal, routing AI-driven itinerary requests through Sabre’s Mosaic APIs for live pricing and availability across more than 420 airlines, with PayPal handling payment execution inside the same conversational flow.
The pattern across both approaches is consistent: no GDS is letting a language model guess at inventory. The competition is over whose deterministic layer AI platforms choose to call, and whose protocol becomes the default connective tissue. Travelport’s bet is that MCP’s open, standardised nature gives it an edge, since any MCP-compliant AI system can integrate with TripServices without custom engineering on either side. Sabre’s approach, at least in the Mindtrip deployment, has run through more bespoke partnership integration.
For distribution executives, the open-protocol argument is worth watching independently of who wins the current round. If MCP becomes the default handshake between conversational AI and booking systems industry-wide, the GDS or NDC aggregator that is easiest to reach through it gains a structural advantage that has nothing to do with content depth or fare coverage.
What this signals for emerging-market distribution
None of the 400 newly connected agencies are outside Europe, and Travelport has not indicated the Orchestra integration extends its reach into Africa or the Middle East. That limits the immediate relevance for TDN’s core audience. But the architecture question does not stay contained to one region for long. NDC aggregators and DMCs across Africa and the Gulf are watching the same AI-agent booking trend, often through the same vendors: Travelport, Sabre, and Amadeus all serve emerging-market agencies alongside their European books of business.
The practical question for operators in these markets is not whether to adopt agentic booking today. It is which GDS or aggregator relationship positions them to plug into a deterministic AI layer without a rebuild, once traveller trust and regulatory clarity catch up to the technology. Travelport’s data point here is worth noting: API-driven transactions have grown from 43% of its customer transactions in 2022 to 63% in 2026, evidence that the shift toward machine-readable, structured booking access is already underway independent of the consumer-facing AI trust question.
The Orchestra deal is a European distribution story with a genuinely global infrastructure argument attached to it. The distribution story will be replicated elsewhere. Travelport and Travelsoft are betting that the deterministic layer, not the conversational one, is where the real distribution advantage will sit. On the evidence so far, that bet looks better placed than the industry’s enthusiasm for AI booking itself.
TDN will continue tracking Travelport’s TripServices rollout and MCP-based agentic booking deployments as they extend beyond Europe.
