Aktiver — semantic AI platform Deterministic · traceable · behind your firewall
Insights / Ontologies

Oracle puts AI-generated business ontologies at the centre of its data platform

At Oracle AI World in Las Vegas on 27 October 2026, Oracle announced new capabilities for its AI Data Platform built around business ontologies: formal definitions of an organisation’s concepts, terminology, relationships and policies that analytics, agents and applications can share. An AI Discovery agent will generate the ontologies from existing systems and documents, and domain experts will review and approve them. Oracle says the capabilities will be available within the next 12 months.

“To scale them reliably, enterprises need a system of context that reflects how the business actually works,” said T.K. Anand, an executive vice president at Oracle, IT Brief reported, referring to AI agents.

/ 01What Oracle announced

The announcement has four main pieces:

  • Business ontologies, drafted by an agent. According to Oracle’s release, the AI Discovery agent analyses database schemas, semantic models, application and API documentation, and knowledge held in unstructured documents. Business experts review and refine the result before approving it, and the platform then monitors how agents use the ontology and recommends updates for expert review.
  • Semantic Reasoning Engine. It uses the ontologies to assemble business-ready context before a request reaches the model, orchestrating retrieval across federated systems and applying governance controls. Oracle says reusing prior analysis cuts repeated reasoning and token use.
  • Deep Insight Agents. Agents that investigate business questions, run forecasts and scenarios, and recommend actions, with human review and approval.
  • AI Registry. A central inventory of agents, tools, skills and MCP servers, with ownership, versions, permissions and lifecycle status. A related AI Governance and Control capability, for testing, policy assessment and monitoring of agent behaviour and cost, is still in development, IT Brief reported.

Analysts quoted by IT Brief framed the move in competitive terms. IDC’s Megha Kumar said “the next competitive advantage in enterprise AI won’t come from models alone,” and BARC’s Kevin Petrie noted that “anyone can adopt frontier models and agentic AI.”

/ 02Why it matters

One of the largest enterprise software vendors is now saying, in a product launch, that agents need an explicit model of the business to work reliably, and that the model should be an ontology. A week earlier Neo4j launched a context product built around a business ontology. The direction of the market is clear: the differentiator is moving from the model to the meaning the model is given.

The workflow Oracle describes is also worth noting: a machine drafts, an expert approves, and the system proposes updates for the expert to review. That is the right division of labour. Building an ontology by hand is slow, which is why most organisations never did it. Letting a model build one unsupervised produces something plausible but untrustworthy. Automation with expert sign-off is the practical middle.

/ 03Where ontologies come from, and where they go

Two questions will decide how useful these ontologies turn out to be. The first is coverage. Oracle’s discovery agent reads schemas, documentation and documents, but much institutional knowledge lives in PDFs, manuals and reports outside any database. An ontology that reflects only the structured estate will miss much of what experts actually reason with.

The second is portability. An ontology defined inside one vendor’s platform is valuable to that platform. An ontology expressed in open standards such as OWL and RDF, and aligned to an upper ontology such as BFO, is valuable to every system that needs to understand the business, including partners’ systems and future tools. The announcement material we reviewed does not say which formats the ontologies will use.

/ 04What to watch

  • Timing. The 12-month availability window, and which pieces ship first.
  • Standards. Whether Oracle’s ontologies import and export in W3C formats.
  • Governance. The still-planned AI Governance and Control capability, which is where testing and monitoring of agent behaviour will sit.
  • Evidence. Whether answers from the Semantic Reasoning Engine carry a traceable path back to the ontology terms and source records that produced them.

Whichever platform is used, the lesson of this launch is that the business model behind the agents is now the part worth investing in.

Aktiver’s take — Manufacturing & energy

Oracle’s draft-then-approve workflow is the right one. Its likely blind spot is the source list: in plants and utilities, the knowledge agents most need sits in manuals, maintenance logs and engineering reports rather than schemas.

Take a turbine fleet. The ERP knows asset IDs and work orders; the procedure that says which vibration reading forces a shutdown lives in a scanned OEM manual and in the notes of a technician near retirement. An ontology linking equipment, failure modes, procedures and inspection records, with each procedure node pointing to its page in the manual, lets an agent answer “which assets are overdue for the inspection this failure mode requires?” and show the clause that makes it so. When a procedure is revised, SMEs update the graph without writing code and every dependent recommendation follows. Built from PDFs as well as SQL and exported to OWL or RDF, that model belongs to the operator, not to whichever platform consumes it.

How we’d model it

EquipmentFailure modeMaintenance procedureManual sectionInspection recordhasFailureModemitigatedBydocumentedInperformedOnEquipmentFailure modeMaintenance procedureManual sectionInspection recordhasFailureModemitigatedBydocumentedInperformedOn
  • Equipment —hasFailureMode→ Failure mode
  • Failure mode —mitigatedBy→ Maintenance procedure
  • Maintenance procedure —documentedIn→ Manual section
  • Inspection record —performedOn→ Equipment

See where this applies →

Go deeper

See how Aktiver turns raw data into knowledge graphs your AI can reason over — deterministic, traceable, behind your firewall.

← All insights Subscribe via RSS → Book a demo →
Keep reading