For most of the past two years the conversation about AI in travel has been about assistance. A chatbot on the booking funnel, a summariser in the reviews section, a copilot for support agents. Useful, occasionally impressive, but bolted onto the side of the business rather than wired into it. That framing is now quietly out of date. The shift happening across booking layers and hospitality operations is towards agentic commerce — systems where AI does not merely answer questions but searches inventory, compares options, assembles itineraries and initiates transactions across supplier and distribution workflows.
Hotel Dive's 2026 trends coverage names agentic commerce as a genuine booking-layer change as consumers increasingly use AI to plan and book. Amadeus reports a 64% yearly increase in AI usage in travel. HospitalityNet and Hospitality Horizons both centre their 2026 outlooks on AI automation as a strategic force rather than an experiment. The interesting story for anyone running a product or engineering organisation is not the technology in the abstract. It is what owning a decisioning layer — one that transacts on a customer's behalf — does to team composition, accountability and the cost of the people you need to hire.
From conversational features to transacting systems
A chatbot that mishandles a query produces an annoyed customer. An agent that books the wrong room, applies the wrong fare rule or commits to a non-refundable rate produces a financial and contractual problem. That difference in consequence is the whole story. Once an AI system can initiate a transaction, the engineering organisation inherits a category of risk it previously left to humans and rules engines. The work stops being about natural language quality and starts being about decisioning, inventory logic and the boundaries within which an agent is allowed to act.
This is why the shift reshapes teams rather than simply adding a feature squad. The companies furthest along — the Booking.com and Expedia tier on the distribution side, Mews and Cloudbeds on the operations side — are discovering that an agent sits across product, data, distribution and trust functions simultaneously. Nobody owns it cleanly, which means somebody senior has to be hired or appointed to own it explicitly. That is a structural change, not a hiring top-up.
Key hiring implications
- AI product leaders who can own a transacting system end to end, not feature owners who treat AI as an add-on
- Engineers comfortable with decisioning and constraint logic rather than purely conversational interfaces
- A clear accountability owner for agent behaviour, usually at director or VP level, given the financial exposure
Workflow automation moves from back office to revenue
Automation in hospitality used to mean the unglamorous middle — channel management, rate distribution, night audit reconciliation. EHL Insights describes broad adoption of AI for personalisation, cost reduction and pricing optimisation, and that operational scope is exactly where agentic systems land first. The difference now is that automation engineers are working on flows that touch revenue directly. An agent that assembles a multi-supplier itinerary is performing the work a revenue manager and a distribution analyst used to split between them.
This pulls workflow automation engineers up the value chain and changes how they are valued. The skill is no longer scripting integrations between known systems. It is designing flows where an autonomous component makes choices inside a transaction — and building the observability to know what it did and why. The people who can do this credibly tend to come from a payments, marketplace or pricing background rather than from a pure integrations background, and they price accordingly.
In-demand technical skills
- Orchestration of multi-step transactional flows with autonomous decision points
- Observability and logging built for non-deterministic systems, not deterministic pipelines
- Familiarity with pricing, inventory and rate logic, ideally from a marketplace or distribution context
Governance becomes a hiring priority, not a compliance afterthought
When a system can commit funds and inventory on a customer's behalf, governance stops being a documentation exercise. The questions become operational and urgent. What is the agent permitted to book without confirmation? How does it handle ambiguity in a fare rule? Who is liable when it gets something wrong, and how is that traced? These are not questions a legal team answers after launch. They have to be designed into the system, which means data and governance specialists need a seat at the table while the architecture is being decided.
For the GDS incumbents — Amadeus, Sabre and Oracle Hospitality on the property side — this is partly familiar territory, because they have always operated under heavy contractual and regulatory constraint. For the faster-moving platforms it is newer and harder, because their culture rewards shipping. The hiring tension this creates is real. A governance specialist who slows down a launch is doing their job; an organisation that hasn't decided whether it values that will struggle to keep one.
Roles gaining importance
- Data governance specialists who understand both model behaviour and travel-specific liability
- Trust and safety engineers focused on agent action boundaries rather than content moderation
- Product owners able to translate contractual fare and rate rules into machine-enforceable constraints
Distribution shifts when the buyer is an agent
If consumers increasingly let AI plan and book — and the Amadeus figure suggests usage is scaling fast — then the entity reading a hotel's content and rates is no longer a human comparing options on a screen. It is an agent parsing structured and unstructured data and making a selection on criteria the supplier may not fully see. That changes how content is structured, how rates are exposed and which API surfaces matter. A hotel group such as IHG, Hilton or Marriott now has to consider whether its inventory is legible to agents at all, not only whether its booking page converts.
This elevates the distribution and partnership technologist — a role that has always existed but rarely commanded the seniority it is now acquiring. The person who understands both the commercial mechanics of distribution and the technical reality of how an agent consumes and acts on supplier data is rare. They sit between the GDS relationships, the direct channel and whatever new agent-facing surfaces emerge, and they are increasingly the people who determine whether a supplier wins or loses the agent-mediated booking.
Roles gaining importance
- Distribution technologists who can reason about agent-readable content and rate exposure
- Partnership engineers managing the interface between supplier inventory and third-party agents
- Commercially literate product people who treat agents as a distribution channel with its own economics
Build, buy, and the vendor question underneath
Few companies will build a full agentic stack from scratch. Most will assemble it from foundation models, orchestration tooling and their own decisioning and data layers. That makes the build-versus-buy boundary a hiring decision as much as a vendor one. A company that buys an off-the-shelf agent layer needs people who can integrate, govern and constrain it. A company that builds needs the rarer and more expensive talent to design the decisioning core. The two paths produce very different team shapes and very different salary bills.
The smaller operators — the Hostelworld and Tripadvisor tier, or independent groups using Mews or Cloudbeds — will mostly consume capability that the larger platforms and vendors expose. Their hiring need is for integration and governance literacy rather than core research talent. The platforms themselves face the harder build, and they are competing for the same small pool of people who have actually shipped an autonomous transacting system. That scarcity is what sets the price, and right now the price is rising faster than the supply.
Key hiring implications
- Buyers need integration and governance depth; builders need scarce decisioning and applied AI talent at a premium
- The pool of people who have shipped a live transacting agent is small, and compensation reflects it
- Team shape follows the build-versus-buy decision, so the org design and the architecture decision are effectively the same conversation
Where this leaves the field
The move to agentic commerce is less a new product category than a redistribution of responsibility. Decisioning that lived in revenue managers and distribution analysts is migrating into systems, and the accountability for those systems has to live somewhere new in the organisation. The roles that gain — AI product leadership, workflow automation, data and governance, distribution technology — are not new titles so much as existing ones being pulled towards the centre of the business and revalued in the process.
What's striking is how unevenly this is landing. The GDS incumbents have the governance instincts but slower delivery cultures. The platforms have the delivery speed but are short on the constraint discipline a transacting system demands. The suppliers are only beginning to ask whether they are even legible to the agents now doing the buying. Each of those positions implies a different team to build and a different set of people to find, and the market for those people is already tighter than the public conversation about AI in travel would suggest.
