Quis coniungit ipsos coniungentes? (Who connects the connectors?): A seamless travel journey in search of a traveler

The other day, I had a call with a company I would rather not name explicitly, which showed me their direct-booking app in ChatGPT. There is actually no need to name the company, as what makes the example relevant here is not the company itself, but what it represents. This, like many B2B platforms, is a layer that sits deep within the hospitality infrastructure, providing value to operators while remaining largely invisible to end users, the travelers. It optimizes, compares, predicts, and personalizes, yet it does so beneath the surface, rarely entering the traveler’s field of perception.

Which is precisely why expecting a traveler to actively connect to such a system, or even know its name, introduces a level of abstraction that feels disconnected from actual behavior. The technology and the impact are real, but the cognitive entry point simply is not.

But I realize I am writing in medias res, so this may not make much sense. Let’s start from the beginning.

For years, our industry has been chasing the idea of a “unified booking journey”. First, through the mythology of the superapp, with WeChat as a distant, and slightly exotic, reference point, and more recently through what I call the “AI as UI phenomenon”, a gradual compression of interfaces into conversation, where planning, comparison, and booking are meant to unfold within a single dialogical layer.

On paper, this trajectory makes total sense. Platforms such as Tripadvisor and Viator are extending into conversational environments, integrating with systems like Claude AI assistant and ChatGPT, allowing users to move from inspiration to selection within the same interaction, browsing options, refining preferences, and eventually approaching the moment of booking*.

* That being said, this is the point where AI providers have already started to encounter the hard edge of travel. Discovery is one thing, transaction is another. The fact that a traveler can ask an AI system where to go, what to compare, which hotel to consider, or which experience might fit a certain intention does not mean that the same traveler is ready to delegate the booking itself, with all its invisible but very real implications, price volatility, cancellation policies, loyalty affiliations, payment flows, refunds, customer support, trust, accountability. Generative AI is extraordinarily powerful at reorganizing discovery, but booking remains a much denser object, because in travel, the transaction is never merely transactional. And the market seems to have understood this faster than the narrative. When reports emerged that OpenAI was scaling back its direct-checkout ambitions in ChatGPT, shares of Expedia, Booking Holdings, and Tripadvisor rose almost immediately.

Groups like Accor are experimenting in this space, embedding their ecosystem into ChatGPT so users can search within their own environment. The same applies to major OTAs such as Booking.com and Expedia, which position themselves as natural endpoints of that conversational flow.

All of this points toward convergence, which is understandable given how fragmented travel planning has always been, spread across search engines, OTAs, reviews, maps, social platforms, and brand websites. AI promises to recombine that fragmentation into something more continuous.

Yet the more I look at it, the more this model seems to assume a very specific, very tech-savvy kind of user.

In its ideal scenario, the traveler moves seamlessly across systems, activates connectors, authorizes integrations, and orchestrates the journey almost like a lightweight systems engineer. That behavior exists, but mostly among power users who are comfortable managing multiple tools, connectors, and agents across different domains.

For most travelers, the direction is almost the opposite. The goal is not to manage complexity better, but to avoid engaging with it altogether.

This is where, in my opinion, the logic of connectors starts to weaken. For a connector to have meaning, the user must already recognize what it connects. Booking.com works because it is already a cognitive shortcut. Accor works when there is some level of brand awareness or loyalty. Even then, activating a connector introduces friction that still needs to be justified.

But to make things worse, when the connector points to an underlying system rather than a recognizable B2C brand, a specific booking engine for example, that friction increases significantly. The expectation that a traveler might connect ChatGPT to the [insert vendor name here] IBE, or to any other layer of the hospitality stack, reflects a misalignment between how the industry thinks and how travelers actually behave.

Travelers do not navigate systems. If anything, they navigate abstractions.

This is where the paradox becomes clearer. The industry operates at the level of infrastructure, while travelers operate at the level of recognition. We build systems, and they reduce them to a few trusted entry points. Yet this connector model implicitly asks end users to understand and activate parts of that infrastructure, even though they have little, if any, reason to do so.

The connector model, in its current form, risks repeating a familiar pattern in which technological possibility is mistaken for actual demand. The trajectory may be logically sound, even elegant, in the way it resolves many of the structural limitations of previous systems. Web3 was framed in similar terms, as the inevitable evolution of the web, capable of addressing most of Web2’s flaws at a foundational level. And yet, despite that coherence, the question remains unresolved: where is the volume of users?

The fact that something can be connected does not mean users will feel the need to connect it. It expands what is possible, certainly, but the question remains whether it creates any real value for the end user, or at least enough to be adopted.

I remain convinced that a more unified booking journey, especially within AI systems, is not only plausible but also directionally correct, and I have been discussing it long before ChatGPT was even launched.

What does not hold, at least for now, is the assumption that users will actively participate in assembling it. Asking the average traveler to connect apps, authorize connectors, understand MCPs, or orchestrate agents presupposes a level of awareness and intent that simply is not there yet. And it is not generational. I teach students in their twenties for several weeks each year, and most of them have no idea what an AI agent is.

Until the system becomes implicit, until connections happen without being asked for, without being exposed, without requiring any form of user literacy, this model will remain structurally ahead of its audience.

So yes, convergence will happen. But not because users learn how to connect the connectors.

Rather, when they no longer have to.