What is an ontology?

An ontology is a model of what exists and how those things relate.

In a company, that means customers, accounts, opportunities, products, tickets, contracts, meetings, owners, workflows, promises, risks, approvals, and the links between them.

A company ontology is the map an intelligent system uses to understand what it is looking at.

The simple definition

The word sounds academic because it is. In AI and knowledge representation, Tom Gruber famously defined an ontology as a specification of a conceptualization: a shared vocabulary for the things inside a domain and the relationships between them. The Semantic Web later formalized this kind of modeling through standards like OWL.

For operators, the useful definition is simpler:

A company ontology says what exists in the business, what each thing means, how things relate, and what actions are possible around those things.

Why companies need one

Most companies do not have one clear map.

They have tools.

The CRM has accounts and opportunities. Slack has decisions. Call transcripts have promises. Linear or Jira has delivery state. Docs have process. Finance has invoices. The company works because humans stitch those fragments together in their heads.

A good operator knows that this account is blocked because the champion left, that this feature request affects a renewal, that this Slack thread changes the next sales action, and that the “source of truth” is not one system but the relationship between systems.

The software usually does not know that.

That is why many AI agents look strong in demos and brittle in production. Tool access is not understanding. Giving an agent HubSpot, Gmail, Slack, and docs is like giving a new employee every keycard on day one with no org chart, no operating model, no memory of what happened, and no sense of what they are allowed to do.

Palantir made this operational

Palantir made the operational version of this idea famous.

In Foundry and AIP, the Ontology is not just a data catalog. Palantir describes it as a decision-centric model for the enterprise: objects, properties, links, actions, functions, logic, and security. Their public framing is close to data, logic, action, and security, with the Ontology acting as the layer where humans and AI systems can understand and act on the same business objects.

That distinction matters.

A dashboard helps people see what is happening.

An ontology lets people and agents decide, act, and learn against a shared model of the company.

For Ortelian, that is the important part. The ontology is not interesting because it sounds technical. It is interesting because agents need a usable model of the company before they can safely work inside it.

What an ontology is not

An ontology is not a nicer wiki, a dashboard, or a pile of integrations.

It is also not the whole AI system. Agents still need memory, knowledge, skills, tools, permissions, evals, traces, approvals, and feedback loops.

The ontology is the base map those systems use.

The map tells agents what the company is. Memory tells them what happened. Knowledge tells them what is currently understood. Skills tell them how work is done. Tools let them act. Permissions and evals keep that action bounded.

Without the map, every workflow asks the agent to rediscover the company from scattered context.

Why this matters for agents

A normal agent can read a customer call and summarize it.

A better agent should know what changed.

If a customer says they will expand if a workflow is supported by Q3, the agent should see more than text. It should understand the account, promise, product limit, owner, approval path, renewal risk, and next safe action.

That is what a company ontology makes possible: it turns scattered information into operating context.

This is the difference between AI on top and AI from within. AI on top automates around the company. AI from within starts by making the company legible to agents.

The ontology is only one layer, but it is the base map. Without it, agents are very confident tourists wandering through your software stack.