Sabre is on an agentic journey, announcing in March it had become an artificial intelligence (AI)-native company.
Garry Wiseman, CTO of Sabre, shed light on the company’s transformation at its Compass event in Los Angeles last week.
“The execution of work is often seen as the long haul in the equation, but with AI, that formula changes,” Wiseman said on stage.
AI is not a trend but the “next paragraph,” he said, adding it stands apart from past technological revolutions such as the internet and mobile transformations.
“This is moving faster than any of those shifts,” he said. “Specifically, it's taken only two years for it to reach what we consider to be mainstream.”
The AI playbooks are still being written, according to Wiseman, who said his product engineering world is now oriented around the technology.
Sabre has used AI assistants to help within the software development lifecycle and is leveraging autonomous agents in areas such as end-to-end planning, development, testing and deployment.
Wiseman said Sabre is facilitating around 70 AI projects being developed by clients and partners. The company has separate AI partnerships with Google and with PayPal and Mindtrip, a PhocusWire Hot 25 Travel Startup for 2025.
Following the event, PhocusWire caught up with Wiseman to dig deeper into what Sabre’s agentic journey means for the company and travel distribution as a whole.
This conversation has been edited for clarity and brevity.
What are you offering to AI-native players as an infrastructure that they can't provide on their own?
What we provide is the ability to shop, book and service all the various travel capabilities for flights and hotels, which is something that Sabre has spent decades perfecting.
We have well over 420 airlines. We have 2 million hotel options and more. We have over 35 car rental companies. We've made that available through APIs that we optimize towards agentic use cases.
Compared to our regular APIs that we've had for quite a long time, that are being used by a lot of the travel agents and OTAs, you think about the metaphor as changing in terms of the interaction model.
Normally with a GUI [graphical user interface] you say your origin destination pair, and you select the dates, you might say which class you can fly and then you hit go. Our shopping engine will fetch a whole bunch of results, and it'll return it in chunks of 20, and you can paginate back and forth.
With a conversation, we can only return maybe four or five … contextually relevant results as a preview with the way that the text is displayed, meaning it'll take in information about prior parts of the conversation, so it's an AI-optimized set of capabilities.
That means the likes of a Mindtrip, for example, they're able to call us in natural language and ask for flights … but they just have to ask it in a natural language question. [They] don't have to learn our APIs.
It's basically a conversational format that they're building things in. We have platforms that know how to respond conversationally, and on top of the shopping side of the house, we also are able to do the booking and servicing.
This is where a lot of the special sauce comes in. There's a huge amount of rules and workflow that happens if you think about exchanges and refunds and all these things that involve the airlines … certain things and maybe regulatory items that we have to take into account, so our backend platform takes care of all of that.
What's the risk of AI-generated outcomes eventually becoming kind of samey?
The scale makes it hard to replicate what we're doing.
The reason why we can respond so quickly with the shopping results is because we get over a billion shops a day. From a technology standpoint, that means that we have all these signals that are coming in that allow us to predict what the next person is going to ask for. So, before you even get requests for that particular origin-destination pair, we will have already picked it up, because we do caching that we put into memory. We're extremely fast relative to most of the platforms out there; we have really high performance. We're able to do things in a consistent manner across all of the airlines, all of the hoteliers, all the current companies.
That's very hard to replicate by anyone else outside of the business today.
We're continuing to innovate to make sure that we're staying ahead. I talked a bit earlier about this notion of being event-proof. What it basically means is that if you have your chatbot and you've told it, “Hey, I'm interested in going to Ibiza this summer, but I can only afford so much in terms of flights.” What can happen is your chatbot can subscribe to one of our services where it'll say, “Hey, let me know if there's flights below X dollars.”
It’ll be a Sabre platform component that the chatbot will connect to, and then once it reaches that certain price level, we can ping your chatbot … through the phone or your interface, and say, “Hey.”
This is the new evolution you'll start to see, probably in the next three to six months, in terms of the chatbots becoming more [proactive].
Today, with Gemini, it'll sometimes say, "Hey, you know, here's your daily preview.” Imagine that's going to get more regular. It'll just ping you at some point here in the day.
Can you share more detail about the 70 AI projects in active development?
Seventy projects are being built against our agentic APIs—so these are the customers actually building lots of different things in what we call the sandbox environment.
They can try things out, play around with it. It ranges from some of the startups that we're working with … it could be travel management companies that are trying to automate certain tasks. Sometimes it's the larger OTAs and airlines that are also trying to build solutions.
It's basically a long list of existing or new companies who are trying to build agentic or AI-driven experiences and they're using our platform as the basis.
If all the GDSs and the metasearch players want to be the intelligence layer for travel, who wins?
Who feeds the metasearches?
We have the source of the information in most cases. This is what our core business is. This is really what we do for a living. We are a B2B company, and we also work very closely with the suppliers to help enable their agentic use cases.
For example, with Virgin Australia. We have just launched a ChatGPT plugin for them. This is something we can do for all [of] the 420 airlines on our marketplace, the GDS.
Likewise, with all the hoteliers. We have these blueprints now that we could deploy for our customers if they want to try out being included in ChatGPT, or whenever Google releases its version of a plugin for Gemini. We're there as a tech provider to help make that happen, so it runs deeper than just the marketplace and the content.
We act as a tech partner for these suppliers.
How do you balance the pressure to deploy all these different features and partnerships while maintaining this massive infrastructure?
It's always a balance.
From an engineering standpoint, we make sure we dedicate time and resources throughout the year where we're making sure that we're doing security patching, maintenance work, upgrades, etc. The teams balance it literally on an annual, quarterly and biweekly basis to figure out what they should work on. They're empowered to make those decisions in terms of where they should focus.
But with a lot of automation that we've put in place, it frees up time for us to do more of the new feature type of development work and less of the maintenance or tech debt type of activities.
Trust has been such a theme here and elsewhere. What are you doing to build trust?
When you look at all of our systems and the APIs, they're effectively the same APIs that we've been using for quite a long time, but we've now evolved to be agentic friendly.
The hundreds of thousands of travel agents that use our APIs today, very securely, and there's no data leakage, etc., it's the same foundation for these agentic experiences.
For instances where we've built chatbots on behalf of our customers, we've been very explicit about training the models in such a way where it's always very deterministic whenever we give you information about flights, hotels or car rentals. That means it will never make up a flight that doesn't exist. It'll never tell you something about a hotel that isn't true.
The other way that we do this is by training the model to be much smaller.
It's very specific towards its travel use cases, and we also train it—if you don't know the answer, say you don't know.
Sounds really simple but actually, you have to train the models to make sure that there's no situation where they will start to hallucinate.
Kurt [Ekert] (Sabre CEO) mentioned that you're devoting 40% of your weekly meetings to discussing AI. What are you talking about?
Number one, our product strategy: How we make sure … we have ways of getting these technologies into our customers' hands, and eventually in front of consumers.
Number two is we think about how internally at Sabre we're going to basically change the organizational structure, change the processes and tools, as to how we become purely AI-native being an existing business that has been around for quite a long time.
Most companies have gone through digital transformations that have been around for a while, which is where they've adopted new tools, new processes and basically, from a people standpoint, have trained everyone differently.
Now with AI, we're going through that transition.
We're using AI tools as assistants for doing pieces of the work. But we're going to shift towards having the teams essentially adopt AI agents that will do most of the parts of the work that they might have done before.
I describe it to my engineers and product managers in that today you might be playing an instrument in the orchestra. In the future, you're going to be a conductor of many agents playing instruments on your behalf.
So, you sort of step back from writing code and instead you're directing these agents to make sure that they're writing code in the right way—to our standards, that are scalable, etc. And then you're doing that in concert with someone else who is defining the product requirements.
It's a very different way of working. But we have to transition to that. It's not because we're trying to reduce the workforce. It's because I know if we don't do this in the next six to 12 months, we're not going to be as effective, perhaps, relative [to] startups or competitors.
*This reporter's attendance at the event was supported by Sabre.