The Structure of Intent
5 Aug 2026
Part I: Intent as the inclusion criterion
The semantic layer has traditionally been described in terms of what it contains: business definitions, ontologies, transformations, mappings, taxonomies, metrics, and other semantic artefacts.
This view is descriptive, but ultimately incomplete. It tells us what is present, but not why.
A more fundamental view is possible.
The semantic layer is not defined by its contents. It is defined by its intent.
More precisely:
The semantic layer is the place where intent is encoded.
It contains precisely those abstractions, definitions, transformations, relationships and constraints that are necessary to satisfy a particular intent.
Intent therefore becomes the inclusion criterion.
An element belongs in the semantic layer not because it is an ontology, a business rule, or a mapping, but because it contributes to the realization of the system’s purpose.
This immediately explains why different experts construct different semantic layers for the same domain. They are not necessarily disagreeing about reality; they are encoding different intents.
The semantic layer is therefore not a representation of the world.
It is a representation of which aspects of the world matter for a given purpose.
In this sense, intent is not merely stored within the semantic layer. It is encoded within it. The semantic layer becomes the operational expression of intent, enabling the system to act coherently toward its objective.
I made a distinction between encoded and enabled.
Those aren’t synonyms.
They’re sequential.
- Intent exists.
- The semantic layer encodes that intent.
- The encoding enables coherent action.
That’s a chain.
And it is helpful to make that distinction because it separates three things that are often collapsed into one.
An organisation can have intent that is never encoded.
It can encode intent poorly.
Or it can encode it well, thereby enabling coherent behaviour.
That’s exactly why two companies with the same stated mission behave completely differently. Their missions are similar, but their semantic layers—the way intent has been encoded into processes, language, definitions, metrics, roles, software, and decision criteria—are radically different.
This is the deeper pattern I am converging on:
Intent cannot act directly on reality. It requires an encoding.
In biology, that’s DNA and regulatory networks.
In software, that’s architecture and code.
In organisations, it’s culture, processes, language and metrics.
In knowledge systems, it’s the semantic layer.