← Our View
Essay

Why your ontology is your moat

There is a lot of talk about ontologies in the AI;DR feeds right now. Snowflake recently released their business ontology and there are countless repositories on Github that help you implement or design them. What’s it all about and why should you care?

Our view is that the ontology is actually your competitive moat. While everyone has the ability to make and experiment faster than ever due to AI, that doesn’t mean how their organisation actually works - how it makes decisions - can keep up with that pace.

What an agent ontology is, and why it matters

An agent ontology is a governed model of what an organisation’s AI agents need to know and what they are allowed to do. It holds four kinds of content:

  • Vocabulary: the business terms, such as customer, account and balance, each with an agreed definition, an owner and a link to the data that implements it.
  • Roles: the human roles that agents act for, how each role decides and the questions it asks.
  • Interactions: how work moves between roles through handoffs, approvals and escalations.
  • Rules: what each agent may see and do, and under what conditions.

A business glossary describes data for people to read. An agent ontology also describes the agents, and it is written to be compiled and enforced.

How it drives a knowledge graph

A knowledge graph presents the organisation’s records as connected things: this customer holds this account, which has this balance. The ontology is the definition the graph is built from. It says which kinds of thing exist, how they relate, which tables and columns hold them, and how records from different systems resolve to the same thing.

The ontology also gives each agent a path through the graph. The path sets where the agent starts, which relationships it may follow, what it may read at each step, which functions it may call and where it must hand off to a person. Anything off the path cannot be reached. A role’s paths, filled with current data, are that role’s digital twin.

Why it is critical

An organisation that deploys agents without one meets four problems:

  • Inconsistent meaning: each agent infers what “active customer” means from column names, so two agents give two answers to the same question.
  • Missing local knowledge: a model knows general business concepts. It does not know which table holds a balance, how a status code is read or which threshold triggers an approval.
  • Weak control: a rule written into a prompt is a request the model may not follow. A rule declared in the ontology is enforced outside the agent, at each function call and in the data platform.
  • No accountability: nobody can trace an answer back to the definitions, data and rules behind it, or to the person who approved them.

The ontology also lets the work compound: the tenth agent reuses the terms, rules and graph that the first one needed.

How can we help

We offer advisory help in defining your ontology and software to support its evolution. For those unfortunate enough to have experienced the design of an ‘enterprise’ data model from the top down before, you will know that it takes for ever, nobody agrees on it, and is never implemented as designed anyway. The same applies to ontologies. The issue is less the identification of the business terms themselves, and more the rules applied to who can access data associated with them and what actions are allowed on each.

In our view, this is best done iteratively, as part of an exercise that defines a new operating model for your company. We do this in cohorts of a few teams and projects at a time. We provide a starter core of the ontology, infer customisations from your critical systems and then test the use of this in real projects by running agents. You see the value right away, contain the blast radius of change and create a trusted foundation upon which to build.

The big idea

So the big idea is simply this: an agent ontology defines what is important for your company and the various elements any agent must consider when attempting an action or answer. This definition, evolved as you evolve is the very definition of your competitive advantage.