Airlines are now under growing pressure to prepare for an agentic AI future while still relying on legacy infrastructure that continues to power critical operations. SunExpress is one of the carriers actively experimenting with generative AI models across its workflows. We spoke with Mustafa Egilmezbilek, CIO at SunExpress, about the most promising GenAI use cases beyond chatbots, the lessons learned from real-world implementation, and the upcoming changes the industry must be ready for.
SunExpress is actively experimenting with GenAI across its workflows
Q: How does SunExpress balance GenAI ambitions with the reality of legacy airline systems?
Mustafa: Legacy systems are simply the reality of the airline industry. They aren’t going away anytime soon, because replacing them is a huge effort involving complex business logic and dependencies. At the same time, GenAI doesn’t naturally fit into legacy environments with their siloed data, accessibility issues, and rigid architectures that make integrations more difficult.
Our approach is building a reasoning layer that sits above the systems of record and treats them as infrastructure. We modernize only what is necessary in core and legacy airline systems to unlock targeted GenAI use cases. The systems of record remain stable, while integration and data layers enable GenAI applications to act as systems of intelligence on top. These applications connect data, support decisions, and orchestrate workflows without disrupting the operational backbone.
Q: Most GenAI use cases in airlines today focus on customer experience, like chatbots. Where do you see the biggest opportunities for the technology to drive value in core business operations?
Mustafa: We started with chatbots as well. It was around 2023 when we had already moved to the cloud. Like many other companies, we began using Copilot-like tools to improve employee productivity and to support customer service use cases. So far, we’ve developed more than 25 generative AI solutions.
But one of the biggest opportunities we see is transforming the software development lifecycle — what we could call Agentic SDLC.
This doesn’t just compress timelines; it changes the cost curve for creating and maintaining software. That directly impacts build-versus-buy decisions. Things that were previously uneconomical to develop in-house now become viable, reducing dependency on vendors and licensing costs.
It also boosts organizational agility. Faster integrations and iterations allow businesses to test commercial hypotheses much more quickly and cheaply. The mindset shifts from “Should we build this?” to “Let’s build it and see.”
Another extremely important area for us is automation, especially since we’re not a very large team. We’ve already implemented robotic process automation extensively, developing RPA solutions for more than 300 tasks and processes. One limitation of traditional RPA was that whenever user intervention was needed, the automation would stop. But now, by inserting AI agents with reasoning capabilities into the process, we can automate some workflows end-to-end.
Beyond generative AI, we also continue to see strong potential in traditional AI and machine learning. We’re now using GenAI capabilities to accelerate ML development itself.
For example, data preparation typically takes 50 to 70 percent of the ML development process. By using generative AI, we can significantly reduce that effort. Another important application is feature engineering. Identifying the right features has a huge impact on the model itself, so finding them gives us strong leverage.
Last but not least, machine learning traditionally requires deep expertise in areas like statistics and probability. In the past, building those capabilities wasn’t always easy because of talent shortages and budget limitations. Now, generative AI lowers the barrier to developing these kinds of solutions.
Q: Should we expect other airlines to modernize faster with GenAI, as they suddenly have tools that can dramatically improve productivity and support in-house software development?
Mustafa: Exactly. And honestly, it’s not just airlines — every industry is going through the same transformation right now.
You know, there was a period over the last couple of years when, for the first time, computer science graduates in the US were struggling to find jobs. But more recently, I noticed that the number of software engineering job postings started increasing dramatically again. In my opinion, one of the reasons is that after seeing the rapid progress in generative AI, CIOs and engineering leaders realized they can now build much more internally with these technologies. And airlines are not an exception here.
Q: To what extent can GenAI help solve the problem of NDC API inconsistencies?
I think there’s definitely potential there. Right now, airlines often return NDC offers with inconsistent ancillary descriptions, fare conditions, or baggage policies. So one possible application is an LLM-based normalization layer that can ingest different offer responses and produce a more canonical representation of the data. In a way, it’s similar to the ETL processes we use in data engineering when bringing data from different sources into a data warehouse. I would describe it as content normalization.
Another promising use case is schema mapping. Airlines are still operating on different NDC versions — for example, we are currently on 19.2. Generative AI could help generate and validate mappings between versions, like translating 19.2 structures into 21.3 or warning when certain fields are not applicable. I think this could significantly accelerate integration development.
And then there’s the natural language side of it. Generative AI is very strong in understanding natural language, so potentially it could help translate complex traveler requests into the proper NDC structure. That could improve both the speed and correctness of the entire flow.
Q: Let’s turn to the dark side of GenAI. What were the biggest pitfalls and challenges you encountered? What lessons have you learned?
Mustafa: One of the biggest challenges with generative AI is the risk of hallucinations due to its probabilistic nature, so we have to be very careful. At the moment, whenever we develop mission-critical systems, we always keep a human in the loop, build guardrails, and, most importantly, try to make workflows as deterministic as possible. What I mean is that when it comes to executing critical actions, an agent calls an API rather than relying on another agent.
At the same time, we’re seeing huge investments in observability and evaluation tools. And as the models themselves continue to improve, I believe we’ll eventually reach a point where hallucinations become negligible.
We also discovered that things may look acceptable in a pilot environment, but once you go live, you face messy realities like data quality issues, reliability gaps, and cost explosions. Some projects turned out to be much more expensive in production than we expected.
And building the solution is only one part of the equation. Generative AI applications require continuous monitoring, evaluation, and improvement, which means dedicated ownership. So organizationally, we started trying to assign business-side owners for AI initiatives who could continuously maintain and develop them. Sometimes it worked, sometimes it didn’t.
But overall, the technology is evolving so quickly that everyone is still experimenting and learning. I believe the companies that will succeed are the ones actively trying to use these new capabilities. Laggers who ignore what’s happening and wait for everything to fully mature may eventually realize they’re already too late.
One of the biggest lessons for us has been: start small, iterate quickly, measure success carefully, and prepare the organization for change. Because even if the technology is ready, cultural transformation is still one of the hardest parts.
Q: In terms of cultural transformation, where do you see the biggest pushback? Is it coming from leadership, employees, or both?
Mustafa: Of course, it can come from all directions. Honestly, I think human transformation is the most important ingredient here — and I’m not saying this as a buzzword, but as someone who has gone through this process.
That’s why we partnered closely with HR and began working on AI literacy programs. What we’ve learned is that the best way to increase adoption is to show people what the technology is in one-on-one sessions. I know that’s not easy to scale, but it’s very effective. You sit next to someone who hesitates or resists using the technology, try to understand what they actually do every day, and then show them how AI can help with repetitive or boring tasks.
My understanding is that it’s not a one-time effort. It’s a continuous process. And given the speed at which technology is evolving, everyone must eventually start adapting. The longer you wait, the harder it becomes.
Q: Do you see a future where airlines bypass online travel agencies and instead run something like MCP on top of their inventory, connecting directly to platforms like ChatGPT, Perplexity AI, or Google’s Gemini for distribution?
Mustafa: I definitely see that potential. Right now, though, adoption is still quite low, and I think the biggest factor is trust. But trust is not a product — it’s a process. As these systems become more reliable, user behavior will gradually change.
Q: Even before autonomous booking becomes mainstream, AI-driven search threatens to dramatically increase the volume of shopping requests. How are airlines preparing for that?
Mustafa: That’s a very good point. Today, a typical look-to-book ratio might be around 1000:1. But with agentic commerce, I’m sure those numbers will spike significantly — maybe to 10,000 or even 20,000 to 1. I think better caching mechanisms and similar technologies can help us deal with this challenge. But we also need to be proactive — not just sit and wait until swarms of AI agents start constantly querying booking systems. So yes, we’re already monitoring this very closely and working with our PSS provider to develop solutions.
Q: Where do you think GenAI and agentic AI will take the airline industry in, say, 10 years?
Mustafa: Honestly, the developments are moving so fast that I’m not even sure what will happen by the end of this year, let alone in 10 years.
But from a technology perspective, I think we may eventually get closer to AGI — artificial general intelligence [a hypothetical type of AI that matches or surpasses human intelligence across cognitive tasks — AltexSoft]. And once we reach that level, the pace of change will accelerate even further. At that point, predicting the future becomes much harder.
What I do expect is a world filled with AI agents and multi-agent orchestration. Many processes will likely be handled autonomously by these systems. It even looks like we may start giving agents identities to measure their performance.
How could this impact airlines? For example, disruption management could be fully automated. Irregular operations may become far less chaotic and costly. At SunExpress, we’re already starting to think about building digital twins of aircraft or operational environments.
On the customer side, passengers will increasingly interact through personal AI travel agents that will communicate directly with airline systems to deliver true hyper-personalization and the best offers for you.
I think conversational interfaces may replace many traditional front ends much sooner than people expect — maybe within the next three to four years. But beyond that, it becomes very difficult to predict.
With 25 years of experience, Liudmyla is a seasoned editor and IT journalist. Over the last five years, she has focused on travel tech, travel payments, and the advancements in NDC implementation.
Want to write an article for our blog? Read our requirements and guidelines to become a contributor.