Objects, Links, Actions
The three primitives every enterprise ontology is built from — and why most AI products skip them.
Abstract
Enterprise data is not an ontology. A working ontology is built from three primitives — objects, links, and actions — that together let an AI system reason about a business the way a human operator does. Get the primitives right and the AI is operational; get them wrong and it is permanently advisory.
Three primitives, in order
Every enterprise ontology that actually carries production AI workloads decomposes into the same three primitives. They are not interchangeable, and they are not optional.
- Objects: the entities a business reasons about — a customer, a shipment, a case, a machine. Each has properties (name, status, owner) and a stable identity that survives across source systems.
- Links: the relationships between objects — this customer placed this order, fulfilled by this truck, dispatched by this driver. The links carry as much meaning as the objects do.
- Actions: the operations that mutate the world — book the truck, refund the customer, escalate the ticket. Actions have preconditions, permissions, side-effects, and consequences the AI must respect.
Why each primitive matters separately
Skip objects and the model talks about your business in generic terms — 'the user' instead of 'the enrolled patient on the Tier 2 plan with three open claims'. Skip links and the model cannot reason about chains of cause and effect across the business. Skip actions and the model can only ever advise — never operate.
The most common failure mode
Teams confuse 'we have a database' with 'we have an ontology'. A database has rows and joins. An ontology has objects, links, and actions — and the meaning attached to each. The meaning is what an LLM needs in order to act safely. Without it, you are wiring a powerful reasoning engine into a schema it cannot interpret.
Rows are not objects. Joins are not links. SQL writes are not actions.
What this means for hiring
The engineers who can stand up an ontology are not the ones who write the fastest queries. They are the ones who can sit with a domain expert for two weeks and come back with the right list of objects, the right links between them, and a short, sharp list of actions the AI is allowed to take. That kind of work is what the StarPlan marketplace is designed to surface.