The fastest way to kill an enterprise technology rollout is to make people learn a new, complicated interface before they can do their jobs.
Travel operations teams know the pattern. A platform arrives with real capability, a training deck and a login.
The agents who are meant to use it already move between a global distribution system (GDS), a mid-office system, email and a row of browser tabs. A new console becomes one more place to log into, one more set of clicks to remember when the queue is backing up. Adoption stalls. The capability sits idle while the problem stays exactly where it always was.
Good software dies this way all the time. The model works in the demo. The benchmark numbers look strong. Six weeks later, the agents have quietly routed around it, back to the workflow they already trust.
Adoption is the test that matters
For agentic artificial intelligence (AI), this is the whole game. A system that can resolve a rebooking in minutes delivers nothing if the agent never opens it. Capability is necessary, but on its own it changes no operational number.
The stakes are real. Corporate travelers now expect the same instant, responsive service they get from every other digital tool they use, and travel management companies are under steady pressure to deliver it while holding margins and headcount flat. Agentic AI is one of the clearest paths to meeting that expectation at scale. A system that agents leave unused captures none of that upside, and it adds licensing cost and change fatigue on top of a demand that is only growing.
So the question for any travel operation buying agentic AI runs deeper than what the system can do. It is whether the people on the floor will reach for it on a busy Tuesday.
The behavior already exists, and it formed outside of work
The habit that drives adoption is already universal. People type a question into a chat window and get an answer back, the same way they do with consumer assistants and the apps on their phones.
Phocuswright found that close to 40% of U.S. travelers used generative AI to plan a trip in 2025, an 11-point jump in a single year. Conversational AI moved from novelty to default faster than almost any interface before it.
When an agent sits down at a service desk, they carry that expectation with them. A blank message field reads as obvious. A dense panel of fields and dropdowns reads as homework. Meeting agents inside the interface they already understand, in the tools they already use, removes the learning curve before it can become a reason to give up.
Put the complexity behind the chat, not in front of the agent
Conversational chat works as the entry point because it asks nothing new of the user. An agent types what they need the way they would tell a colleague: Rebook this passenger on the next flight in the same cabin, or cancel and refund the ticket inside the void window. The words are ordinary. The work behind them is not.
The real engineering lives in the gap between that plain request and the booking system. Behind the message field, the system reads intent, checks the traveler profile and the client's policy, runs the multistep GDS sequence and reports back what it found—making sure agents have everything they need to execute the request. The agent sees a short exchange. Underneath, a sequence that used to take a trained person several minutes and a dozen keystrokes runs on its own.
The strongest agentic systems built for travel operations work this way, sitting inside the systems agents already use rather than beside them as a separate product.
Some workflows run automatically, reading the request, validating the details and drafting a response for the agent to approve. Others need agent input—what to look for next, if there are any complexities to address or more. The chat is what the agent sees. The orchestration across GDS and midoffice is what makes the chat worth opening.
Adoption first, then the numbers follow
Once agents use a system because reaching for it is effortless, the operational results show up on their own.
Agentic AI applied to routine service work has the potential to cut handling time substantially, take a large share of repetitive ticket handling off agents' plates and bring response times down from hours to seconds across every time zone.
Every one of those gains depends on agents actually using the tool, which is the entire point of designing it around a habit they already have.
Familiar on the surface, governed underneath
Meeting agents in the chat window does not mean loosening control, and for travel operations it cannot. The system logs every action it takes and keeps a full audit trail. The models run in private environments, and no traveler data goes to a third party.
Escalation thresholds and policy logic sit beneath the conversation, and the system applies them on every request, so an agent never has to carry them in their head. Ease of use at the surface and tight governance underneath come from the same design decision.
Phocuswright's research now puts more than 60% of travel businesses experimenting with or scaling agentic AI. Over the next year, the companies that pull ahead will be the ones whose agents reach for the technology without being asked, and the interface is where that difference gets decided.
Build the entry point as the chat window people already know how to use, keep the heavy lifting behind it and the technology finally earns its place in the operation.
About the author...
Ami Goldenberg is the CTO and co-founder of
Oversee.