Agentic AI in Hospitality and Tourism: A Typology, Six Risk Domains, and a Research Agenda

Most AI systems currently operating in hospitality and tourism respond to instructions, follow rules, or generate content on request. Agentic AI does something categorically different: it pursues goals autonomously, reasons across contexts, adapts to changing conditions, and can coordinate actions across interconnected service systems with minimal human oversight. This conceptual paper maps the emergence of agentic AI in hospitality and tourism, proposes a five-type functional typology, and examines the risks and governance challenges that responsible adoption requires.

How the research was done

The paper is a conceptual contribution grounded in two complementary theoretical frameworks: socio-technical systems theory, which emphasises the interdependence of social and technological subsystems and treats sustainable change as requiring their mutual adaptation; and actor-network theory, which positions non-human entities — including AI systems — as active participants that reshape practices, relationships, and service dynamics. The typology proposed in the paper was developed through a systematic narrative review of scholarly and industry literature, tracing the trajectory from rule-based AI through generative AI to agentic AI in hospitality and tourism. Industry cases including Air Canada's refund chatbot, McDonald's drive-through AI, Marriott's personalisation architecture, and Carnival Cruise Line's chatbot deployment anchor the conceptual argument in documented real-world evidence.

What the research found

The paper proposes five functional categories of agentic AI in hospitality and tourism. Service agents interact directly with guests, personalising service encounters and using natural language processing and sentiment analysis to interpret needs and act autonomously. Planning agents manage dynamic, context-sensitive itineraries, reconfiguring schedules in response to disruptions without requiring guests to re-engage. Monitoring and compliance agents operate in the background, tracking safety standards, sustainability metrics, and service quality through IoT sensors and computer vision. Engagement and experience agents use augmented and virtual reality alongside affective computing to create emotionally responsive guest experiences. Meta-agents function as intelligent orchestration layers, coordinating actions across housekeeping, dining, reservations, and maintenance simultaneously, detecting disruptions and initiating corrective actions without human intervention.

Alongside these opportunities, six interconnected risk domains are identified. Privacy and surveillance concerns arise when agents rely on biometric data and real-time emotion tracking. Algorithmic bias and discrimination emerge when systems trained on unrepresentative datasets make culturally inappropriate or systematically inequitable decisions — a risk illustrated by Amazon's recruitment algorithm, which systematically downgraded female applicants. Explainability and accountability are undermined when agentic systems function as black boxes, as the Air Canada case demonstrated when the airline was held legally liable for its chatbot's false refund promises. Labour reconfiguration and workforce displacement represent a genuine risk, particularly given that most large-scale agentic AI deployments are motivated primarily by cost reduction. Operational fragility and legal ambiguity complete the picture. The paper makes explicit that each risk domain links directly to specific agent categories — the risks are not generic but arise from how particular systems are embedded in particular service contexts.

Insights for the industry

Agentic AI does not simply upgrade existing automation — it redistributes agency and accountability across entire service ecosystems. The failure cases reviewed in the paper make clear that the commercial and reputational costs of poorly governed deployment are concrete rather than theoretical. Air Canada faced legal liability when its chatbot provided false refund policy information. McDonald's discontinued its AI drive-through experiment after the system mishandled accents and complex orders. For hospitality organisations exploring agentic AI, the implication is that governance frameworks — human-in-the-loop oversight, transparent decision architectures, regular bias audits, and clear liability protocols — need to be embedded from the outset rather than added after problems emerge.

The paper's warning about the primary motivation for most large-scale agentic AI deployments also warrants attention: cost reduction, not workforce augmentation, tends to drive the decision in practice, which risks accelerating the displacement of frontline workers rather than productively redefining their roles. Carnival Cruise Line's model — chatbots handling routine booking inquiries, staff freed to focus on high-touch guest experience design — represents a more constructive template and a more honest representation of what human-AI collaboration can look like when it is designed with care. The research agenda outlined in the paper identifies consumer trust, cultural adaptation of AI behaviour, labour dynamics, and long-term societal impact as the empirical priorities most urgently requiring investigation.