Why AI agents are not the new SaaS, and what that means for hotels and tourism

Over the past few months we've been reading the same promise everywhere, dressed up in slightly different clothes each time: AI agents will replace software, revolutionise every company, become the new SaaS and spawn a new generation of unicorns.

It's a compelling narrative, easy to tell and perfect for pitches, conferences and LinkedIn. The problem is that, in practice, things look considerably less tidy.

The intuition I'm starting from today — and one I find myself agreeing with quite a lot — is this: agents are not SaaS. Or at least, not in the sense SaaS has been for the past twenty years.

The reason is simple. SaaS scaled because companies standardised huge numbers of processes around replicable software. AI agents, on the other hand, seem to work best when operating inside processes that are full of exceptions, context, relationships and micro organisational decisions. In other words, rather than universal products, they often look more like operating systems built around a single company.

And this is where the topic becomes particularly interesting for anyone working in hotels, hospitality, destinations and tourism consulting.

SaaS won because work was standardisable

The SaaS model was enormously powerful because it arrived at a historical moment when companies were looking for scalable tools to manage fairly standard functions: accounting, CRM, ticketing, newsletters, HR, marketing automation, project management.

In that world it made sense to think in terms of product:

• many similar customers;
• broadly similar problems;
• same interface;
• same licensing logic;
• low marginal costs as the software grew.

With AI agents — especially the genuinely useful ones — the situation is more complicated. McKinsey, writing about the "agentic organization", insists that value emerges when agents are embedded in real workflows, proprietary data and specific operational processes, not simply bolted on as superficial features to existing tools.

Deloitte says something very similar: many agentic implementations fail not because the technology is weak, but because companies try to graft agents onto organisations, data and processes that aren't ready, or that remain too fragmented.

That completely changes how value is generated.

Agents look more like websites than universally replicable software

The most useful metaphor, in my view, is precisely this: agents will be closer to websites than to SaaS apps.

Almost every serious company, at some point, needs its own digital presence, built around its own identity, its own processes, its own content. But not all of them develop proprietary software.

With agents we could see something similar:

• almost every company will end up having agents;
• but many of those agents won't become venture-scale businesses;
• they'll become bespoke operating systems, built around the way that single organisation actually works.

McKinsey explicitly describes a "mesh" architecture, in which custom agents and off-the-shelf tools coexist and interweave with company data, rules and governance.

For tourism, this idea is far more realistic than the fantasy of a "universal hotel copilot" that works identically in a 400-room resort, an independent boutique hotel, a DMC (destination management company) or a campsite.

The five layers of agents, from simplest to most defensible

L1. Personal agents

These are the ones we see most often today:

• email summaries;
• meeting notes;
• content research;
• writing support;
• personal CRM micro-workflows.

They're easy to understand, relatively simple to implement and often very useful at an individual level. But precisely for that reason they tend to be hard to defend over time. Once a workflow becomes obvious, it also becomes easy to copy. Many will end up as features, not standalone companies.

This dynamic is already fairly visible in the market. McKinsey and Deloitte both agree that the first agentic use cases to be widely adopted concern individual productivity, but they're not necessarily the most transformative on a competitive level.

In tourism, the equivalent is obvious:

• the agent that helps draft a reply to a guest;
• the agent that synthesises reviews;
• the agent that turns internal notes into procedures;
• the agent that prepares a first draft of proposals or captions.

Extremely useful, yes. But hardly a lasting competitive advantage.

L2. Team or function agents

This is where we start moving up a level. We're talking about agents that serve an entire team — for example:

• front office;
• revenue;
• marketing;
• customer care;
• sales;
• administration.

These agents need to know internal documents, policies, exceptions, tone of voice, decision-making hierarchies, the CRM, the PMS (property management system), availability, rate categories and commercial conditions.

Here the value grows — but so does the difficulty.

A good prompt is no longer enough. You need integrations, cleaner data and a genuine process logic.

In tourism I'm already seeing sensible examples:

• agents that help the front office handle complex requests in multiple languages, drawing on FAQs, policies, guest preferences and real-time availability;
• agents that support revenue managers in reading pickup, cancellations, events and anomalous fluctuations;
• marketing agents that coordinate website content, newsletters, social media and paid campaigns while maintaining a degree of coherence.

This is where things start getting genuinely interesting for hotels and groups — but we're still not in the territory of real structural transformation.

L3. Cross-functional agents

At this level the agent no longer serves just one team; it cuts across multiple departments. This is where the game gets serious.

Think of an agent connecting:

• booking;
• pre-stay;
• upsell;
• customer care;
• internal operations;
• post-stay.

Or, in a DMO (destination management organisation) or destination network, an agent connecting:

• destination content;
• transport data;
• events;
• ticketing;
• local welcome;
• customer assistance.

This kind of agent is much harder to build, because it runs straight into a truth we know well: real organisations are full of silos.

And it's also why the value here becomes more defensible. If an agent genuinely manages to work well across departments, it's no longer simply "useful software" — it's a component of how the company functions.

L4. Operational orchestration agents

Here agents begin to behave almost like structured operational work. They're not just doing assistance or support. They coordinate flows, manage exceptions, prepare decisions, route tasks, check data consistency, keep a process alive.

Deloitte explicitly speaks of agents as a "silicon-based workforce" — a workforce that needs to be managed with governance, control and accountability models very different from those of ordinary software.

In hospitality, a level like this could mean:

• a system that orchestrates guest requests, housekeeping, maintenance, concierge and revenue logic coherently;
• an engine connecting pricing, inventory, upsell, staffing and forecast demand;
• an operational machine that doesn't replace management, but reduces repetitive coordination work.

Here agents really do start to look more like "work" than software.

L5. Agents as organisational infrastructure

This is the most interesting level — and, at the same time, the hardest to picture clearly today. We're no longer talking about agents as tools, but as infrastructure deeply integrated into the way the organisation thinks, decides, operates and adapts.

At that point, the agent is no longer "a product". It's part of the company.

And here the comparison with the website or the ERP becomes perfect: you don't just buy it — you design it, shape it, govern it, evolve it.

McKinsey describes this as a new operating model, in which the real challenge is not technical but organisational.

And it's also the point at which many AI startups stop looking "scalable" in the traditional SaaS sense. Because the value shifts from the standard subscription to the project, the integration, the customisation, the understanding of context.

Why all this matters enormously for hospitality

Here I get to the part that interests me most.

In tourism and hospitality, processes are far less standardised than we often like to pretend. Even in large chains, the real work is full of:

• exceptions;
• seasonality;
• territorial differences;
• human relationships;
• operational constraints;
• systems that don't talk to each other well;
• decisions that depend on context, judgement, margins, reputation and timing.

That's why I find it hard to believe the narrative that a single "plug-and-play" agent will change everything in exactly the same way for everyone.

Far more realistic to imagine that, in hospitality, the useful agents will be those built inside and around the real workings of individual businesses:

• in the boutique hotel that wants to improve pre-stay, upsell and customer care without losing its human tone;
• in the hotel group that wants to orchestrate distribution, revenue and operations more effectively;
• in the DMO that wants to become a genuine information and service layer, not just a promotional one;
• in the consultant or marketing team that wants to integrate analysis, content, CRM and campaigns without multiplying manual tasks.

What this means for startups, AI agencies and internal teams

The point, then, is understanding where value accumulates.

If you're building an AI startup

The question isn't only "how good is my agent", but "how deep is the operational context I'm managing to capture".

If you stay too surface-level, you risk becoming a feature. If you genuinely get inside the process, you become harder to replace.

If you're running an AI agency

The business will probably look less like classic SaaS and more like a combination of:

• services;
• integration;
• workflow design;
• governance;
• ongoing maintenance.

That's not a flaw. It's just a different kind of economics.

If you're implementing agents internally

The real question is: which processes actually deserve a bespoke agent, and which don't?

Because not everything needs to be "agentified". But some parts of the organisation definitely do.

The easiest thing to build today is often the least defensible

The simplest agents to build are often the ones that generate the least lasting value.

It's a paradox typical of this phase of the market:

• what impresses most today often becomes commodity soon enough;
• what requires more organisational work today might be the real advantage of tomorrow.

In tourism, that means it won't be the mini assistant writing captions or summarising emails that genuinely shifts the competitive position of a brand or a destination.

The real movers will be the systems that manage to reduce friction between departments, improve the coherence of the experience, get information flowing better and support faster, more contextual decisions.

In conclusion, AI agents are not the new SaaS. Or rather, not only.

In many cases they'll be personal tools, small utilities, features that will improve productivity and convenience. But the real structural value — especially in hospitality — will emerge when we stop thinking of them as "smart apps" and start seeing them for what they actually are: operational work systems, deeply tied to the context, data and real processes of an organisation.

And that's precisely what makes them harder to build — but also more interesting.

Because if SaaS won by standardising work, agents will win, perhaps, where work cannot be fully standardised. And in tourism — which runs on complexity, exceptions and people — that's not bad news. It's simply an invitation to stop looking for the universal shortcut and start designing systems that genuinely resemble the way we actually work.