Why ontologies matter, why they fail and how to build them anyway

2026-06-17 · Source: Thoughtworks Insights · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Project & Product Management · Depth: Intermediate, medium

Summary

Enterprise AI's evolution beyond simple chatbots and vector search necessitates robust semantic understanding for agents orchestrating workflows and acting on business behalf. Published June 17, 2026, this article explains that while language models lack operational logic, domain ontologies can bridge this gap by defining terms and relationships. It clarifies that ontologies are blueprints, distinct from knowledge graphs which populate these blueprints with data. The piece outlines a continuum from schemas to knowledge graphs, emphasizing that full ontologies are justified for multi-hop reasoning and automated constraint checking. Common failures include the "translation gap," "scope creep," and the significant "maintenance problem." To succeed, the article advocates treating ontologies as products: scoping to specific use cases, tying concepts to funded business questions, using AI for drafting, assigning clear ownership, and establishing a lifecycle. This product approach contrasts with conventional methods, promising value in 4-8 weeks per iteration versus 12-24 months.

Key takeaway

For AI Architects evaluating semantic solutions, recognize that full ontologies are not always necessary; prioritize the minimum semantic structure for your problem. You should treat your ontology as a product, scoping it to a high-return use case and establishing clear ownership and a lifecycle for continuous maintenance. This approach ensures value delivery within 4-8 weeks, avoiding the common pitfalls of enterprise-wide, project-based initiatives that often fail due to scope creep and maintenance issues.

Key insights

Treating ontologies as products, scoped to use cases and continuously maintained, is crucial for successful enterprise AI semantic understanding.

Principles

• Ontologies are blueprints; knowledge graphs populate them.
• Formal expressiveness adds cost and governance weight.
• Maintenance, not construction, is the primary challenge.

Method

Build minimum viable ontologies for high-return use cases, using LLMs to draft taxonomies. Fund concepts based on competency questions, then assign ownership and a lifecycle for continuous reconciliation.

In practice

• Scope ontology to contract compliance or onboarding.
• Use LLM pipelines to draft taxonomies from existing data.
• Assign a product owner for ontology lifecycle management.

Topics

• Enterprise AI
• Ontologies
• Knowledge Graphs
• Semantic Layers
• Data Governance
• LLM Pipelines

Best for: AI Architect, Director of AI/ML, Consultant

Related on AIssential

• Ontologies Are the Intentional Core of a True Knowledge Graph — Formal ontologies provide the stable, intentional semantic backbone essential for reliable enterprise AI and knowledge graphs.
• Semantic Foundations for Reliable Enterprise AI — Semantic clarity in data, achieved through robust data contracts and ontologies, is crucial for reliable enterprise AI outputs.
• What Do You Need to Create a Useful Ontology? — Building effective ontologies requires clear scope, purpose-driven design, and careful selection of semantic tools.
• Ontology, Part IV — Ontology governance is crucial for maintaining coherence and preventing drift in deployed ontologies.
• Ontology, Part II — Effective ontology design requires structured thinking, competency questions, design heuristics, and a three-box architecture.
• The Ontology Engineering Challenge is Not Building It, but Mirroring the Delta — The core challenge in enterprise AI is correcting LLM's inherent concept understanding where it diverges from specific business definitions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Thoughtworks Insights.