I’ve been stuck on a question that sounds abstract until you try to build something with it: does the perceptual metaphor initially chosen for a technology determine its evolutionary trajectory?
When a technological category is born, most of the time what we define as "the product" is more the representation of the underlying mechanism than the lower level itself.
For you as a user, your phone is the object not how it works, like the computer is not the transistor, the software is not the code, and the Internet is not the TCP/IP protocol.
For most people, the category coincides with its dominant interface. To make an analogy with Kant (not essential, but it does make the concept embarrassingly intuitive ;)), the interface has always been the phenomenal representation of the product, designed with the aim of defining the modalities of interaction with the object and the metaphor to which to associate them.
The dominant phenomenal representation of a technology does not merely make it usable: it unequivocally brings along the perceptual metaphor it evokes. And that metaphor — to which you associate the interface and consequently the product — does not only influence marketing or the initial UX. It influences what users expect, what designers consider natural, which problems seem important, and which directions seem "obvious". In essence, it tends to orient its future evolution, generating a true cultural inertia.
One naturally wonders: how would certain product categories have evolved if we had changed the perceptual metaphor initially chosen for the design of the interface?
It may seem counterintuitive, but the moment you associate an interface with a mental image, you identify the product within the interface itself. Consequently, the product expresses itself through that metaphor, and its natural evolution, by inertia, will tend to accommodate the expectations linked to that image.
If the content within a computer is presented as a stacking of pages, the user will expect to be able to turn the page, organize them into archives or libraries, place a bookmark, or close the sheet and put it away. The metaphor of the page clearly refers to the card-based computers of the past: not because it was the only possibility, but because the initial perceptual metaphor generated a specific inertia.
But what if they had been boxes? One would expect to be able to stack them, a concept of a "full container" would emerge (an evolution towards the concept of containers), with objects of different typologies inside, not just pages, with entirely different shapes, sizes, and interactions with the screen space.
Or again: how would the concept of social networks have evolved if the feed, instead of evoking an infinite page of multimedia content (as if it were a photo and video album), had chosen a different metaphor, like a large tree? How radically would the architecture of UI flows have changed, along with how users perceive the application, the associated mental images, and, consequently, what naturally becomes expected as an evolution?
No one questions the metaphor of the page or the window. It is so deeply rooted that it seems inevitable, but this point of reflection serves to remind us that it was not inevitable at all. And here the question shifts from whether to how much the initial metaphor determines the evolutionary trajectory.
If representation influences the direction of development, what metaphor are we choosing today for agents?
Software as an Invisible Infrastructure
It rarely happens that products do not have a phenomenal representation to relate to; one never interfaces directly with the noumenon. But when a new category is born, especially if derived from existing products, the definition becomes more blurred: to what extent are we using the previous product? Is this new paradigm truly definable as a "product", or is it rather an infrastructure? If it can already lean on the phenomenal representation it derives from, does it make sense to design one custom-tailored to it?
The answer to the last question is a resounding yes, precisely because of how determining the perceptual metaphor is for future developments.
Let's try to apply this reflection to a concrete context: AI and MCP (Model Context Protocol).
The moment the value — and consequently the product — is no longer a static functionality (e.g., "this program cuts videos"), but is a dynamic workflow (the AI using MCP to connect Unity, Git, and the file system to create a game for you), are we selling the process or the output?
Today, many people are still trying to represent AI to themselves as traditional software: they open ChatGPT just as they would open Photoshop or Excel. But if you take MCP, agents, tool calling, and service orchestration seriously, the product is no longer the software.
The product becomes:
a capability,
a result,
a transformation of the state of the world.
The user no longer uses fifteen different pieces of software: they express an intention, and the AI decides how to achieve it. In this sense, software is effectively becoming an invisible infrastructure.
The Paradox of the Perceptual Black Box
If the user cannot see the available tools, does not know the functionalities, and does not know what is possible to do, how can a perception of the product be formed? (Leaving aside for a moment the purely UX-related implications, which would deserve an examination of their own).
Historically, this situation is rare. In the GUI you saw the buttons, in a car you see the steering wheel, in Photoshop you see the tools: the interface itself taught the possibilities.
Moving towards a world of "workflow as a product", does the conversational interface perhaps become our raw way of relating directly to the noumenon, without any intermediary phenomenal representation? Does it mean that today we use the product without seeing it, in the most literal sense possible, without even a minimal surface (an object, an icon, a body) upon which to land our perception?
This generates a paradox: the AI is perfectly aware of the tools at its disposal and is autonomous in choosing or proposing the right tool to achieve a result; the user, however, not being aware of it, cannot ask for something they do not know they can do. It is a black box. The AI is perfectly capable of executing anything you ask, but does it make sense for there to be no interface proper to this specific transition in the level of abstraction?
The "naked-chat" we use today risks being simply the equivalent of the terminal before applications. Not because it lacks an interface adequate to the noumenon (the underlying layers we interact with are already phenomena, even if they become the noumenon of the product built on top of them), but because a phenomenal identity built over this infrastructure has not yet emerged.
Which Metaphor for the Future?
It makes sense to wonder how the phenomenal representation of this new category of products should be thought out, considering that we will tend to identify the product itself within its representation.
What if the phenomenal representation were shifting from the set of tools to the agent that utilizes them?
If representation influences evolution, then the way we depict AI agents in the coming years will profoundly condition their trajectory:
If we represent them as chatbots, they will evolve toward better chatbots.
If we represent them as personal assistants, they will evolve toward assistants.
If we represent them as ambient operating systems, they will evolve toward invisible infrastructures integrated into space.
The metaphor chosen to represent the noumenon is not neutral. It determines what society will expect from that technology and the direction toward which it will be pushed by inertia.
When a new technological category is born, what we call the "product" rarely coincides with its internal mechanism. Because of this, the most interesting questions to ask ourselves today are: what is the right metaphor to represent a system composed of hundreds of invisible tools orchestrated by an intelligence that acts on intentions rather than commands?
And above all: which initial metaphor risks trapping agents in the wrong trajectory for the next twenty years?