Why Scaling AI Agents Is Harder Than Building Them
Moving travel AI agents from pilots to production requires navigating complex infrastructure, legacy systems, and specialized industry talent.
Stock photo for illustration only, not from the actual event
- Scaling AI agents relies heavily on surrounding infrastructure, systems, and talent rather than just the model.
- Travel companies must choose between building proprietary capabilities or partnering with specialists.
- The agent iceberg illustrates that foundational backend systems outweigh the visible LLM layer.
- Travel sector challenges include legacy mainframes, demand volatility, and multilingual customer support requirements.
Scaling artificial intelligence agents in the travel sector is less about the underlying model and much more about the surrounding systems, talent, and infrastructure. As travel companies transition from initial pilots into full production environments, leadership must carefully determine which technological capabilities differentiate their business and which functions are better delegated to a specialized partner.
While AI agents can appear remarkably proficient in a controlled demonstration environment, deploying that same technology in front of real customers connected to live systems introduces a completely different scale of complexity. Enterprise AI agents must integrate seamlessly with legacy architecture, respond fast enough to maintain natural conversational flows, absorb sudden demand surges, and strictly enforce brand guidelines at scale.
Stock photo for illustration only, not from the actual event
Analyzing the agent iceberg concept reveals that deploying conversational AI is primarily an engineering and integration challenge. Maintenance requires keeping pace with rapid foundation model releases, making the decision to build in-house versus partnering a critical strategic crossroads for travel tech leaders.
Deploying production systems typically follows an incremental path, starting with a narrow scope handling a small fraction of channel traffic before expanding. Sierra utilizes the concept of the agent iceberg to demonstrate how much operational work sits beneath the surface. Above the waterline lie large language models, retrieval-augmented generation, and system tools, while below the surface sit core system integrations, runtime governance, context management, and continuous optimization.
"We hear this a lot: It's just an LLM call with some tools and RAG. The reality is companies tend to underestimate what's under the hood."
Zahnlecker
Three operational realities are especially pronounced within the travel industry. First, legacy technology remains a hurdle as some core airline mainframes date back decades. Second, volatility from weather events like severe winter storms can instantly ground hundreds of flights, overwhelming human support teams while scalable AI agents absorb the traffic spike. Third, language diversity requires global airlines to support dozens of locales, such as Philippine Airlines utilizing Sierra to switch fluidly between Tagalog, Taglish, and English within a single conversation.
Sustaining these complex deployments also depends heavily on internal workforce capabilities. Data from Skift Research’s State of Travel 2026 report indicates that travel maintains one of the lowest proportions of AI-trained employees among major global industries, with only 3% of the workforce possessing formal AI training.
Source: Skift
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment