Every few years, hospitality gets a new word to put in front of “AI.” We’ve had smart, predictive, generative. Now it’s agentic — systems that don’t wait to be asked, but act: setting a room rate at 2am without a revenue manager’s sign-off, rerouting a guest complaint to a refund before a human sees it, reordering stock because a forecasting model decided demand had shifted. The pitch is seductive: less friction, faster response, decisions made at a speed no human team could match.
Here at ISEH, we spend a lot of time asking students to separate what a technology can do from what it should do, and from what it’s actually doing today rather than in the vendor’s roadmap slide. Agentic AI is a good test case for all three, because the gap between them is wider than the current wave of enthusiasm suggests.
WHAT’S GENUINELY NEW HERE
To be fair to the hype: something has changed. A chatbot answering “what time is checkout” is trivial and has been for years. What’s different now is systems making multi-step operational decisions — dynamic pricing that adjusts continuously against live demand signals, personalization engines that decide what to offer a guest before the guest asks, scheduling and inventory systems that act on forecasts without a manager in the loop. That’s a real shift from automation-as-tool to automation-as-actor, and large hotel groups are already running it in production rather than treating it as an innovation-lab experiment.
The operational case is genuinely strong in a specific, narrow band of decisions: high-frequency, low-stakes-per-decision, data-rich environments. Room pricing across thousands of nights is exactly that. So is inventory reordering. These are decisions where being wrong 5% of the time and right 95% of the time is a clear net win, and where a human doing the same job manually would be slower and less consistent anyway.
WHERE THE CRITICAL EYE NEEDS TO OPEN
That’s also precisely where the argument gets weaker as you move it toward the guest-facing, high-stakes, low-frequency end of hospitality — which, inconveniently, is most of what makes hospitality hospitality.
Three things worth sitting with before anyone puts “agentic AI” in a strategy deck as though it’s simply a stronger version of what came before:
“The AI decided” is not an answer that survives contact with a guest relations crisis.
Accountability doesn’t automate. When a system autonomously waives a fee, denies a refund, or reprices a suite the night before a wedding party arrives, someone still has to own that decision when a guest complains, a journalist calls, or a regulator asks a question. Most organisations deploying these systems have not built the escalation and liability structures to match the autonomy they’ve handed over. The technology has moved faster than the governance.
The experience economy and the decision economy are pulling in opposite directions. We’ve spent this same season telling students that guests increasingly want immersive, human-centred experiences over standardized service. Agentic AI’s strongest use case is standardization at scale — exactly the thing the experience-economy trend says guests are moving away from. Both trends can be true simultaneously, but only if hospitality leaders are precise about where they let the machine decide (pricing, logistics, inventory — invisible to the guest) and where they deliberately keep a human in the loop (anything the guest actually notices as a decision made about them). Blurring that line is where brand trust gets damaged quietly, one autonomous decision at a time, long before anyone notices the pattern.
“Agentic” is doing some marketing work, not just describing an architecture. A meaningful share of what’s being sold as agentic AI right now is a chatbot with a slightly longer decision chain and a confident name. Before any organisation treats this as a settled operational category, it’s worth asking a vendor plainly: what decisions does this system make without human review, what’s the error rate on those decisions, and what happens when it’s wrong. If those answers aren’t crisp, the “agentic” label is aspirational, not descriptive.
WHAT THIS MEANS FOR HOW WE TRAIN PEOPLE
The honest implication for hospitality education is that the job that’s disappearing isn’t the front-desk or revenue-management role — it’s the role of not questioning the system. The skill that becomes more valuable, not less, is the judgement to know which decisions are safe to hand to a system optimising for throughput, and which decisions need a human who can be held accountable and who understands that a guest is not a data point mid-negotiation.
That’s not a reason to be sceptical of the technology for its own sake — the pricing and logistics gains are real and worth capturing. It’s a reason to be sceptical of the framing that treats “more autonomy” as an unambiguous upgrade. In an industry whose entire value proposition is built on someone noticing what a guest needs before they ask, handing that noticing over to a system that optimises for statistical patterns rather than the specific person in front of it is not a small technical choice. It’s a strategic one, and it deserves to be treated as a strategic one — not adopted because a competitor announced it first.
ABOUT THE AUTHOR
Erik Lindqvist is Lecturer in Digital Innovation at the Institute for Sport, Events & Hospitality (ISEH), Geneva. His work focuses on where automation genuinely improves hospitality operations — and where it quietly erodes the things guests actually value.
