# Ontologies, Context Graphs & Semantic Layers — What AI Actually Needs

**Author:** [Jessica Talisman, MLS](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fwww.linkedin.com%2Fin%2Fjmtalisman%2F%23this) — Information Architect, Semantic Strategist
**Publication:** [Metadata Weekly](https://metadataweekly.substack.com/) by Prukalpa Sankar
**Source:** https://metadataweekly.substack.com/p/ontologies-context-graphs-and-semantic
**Published:** January 22, 2026
**RDF:** [Turtle](../rdf/ontologies-context-graphs-semantic-jessica-talisman-deepseek_v4pro-1.ttl) | [JSON-LD](../rdf/ontologies-context-graphs-semantic-jessica-talisman-deepseek_v4pro-1.jsonld)

---

## Overview

This knowledge graph captures [Jessica Talisman](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fwww.linkedin.com%2Fin%2Fjmtalisman%2F%23this)'s January 2026 Metadata Weekly analysis of the architectural divergence between [semantic layers](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23semanticLayer) (metrics-first, designed for human BI consumption) and [ontologies](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23ontologyConcept) (meaning-first, supporting logical inference). Talisman traces the history from LookML (2012) through Palantir Foundry to dbt's MetricFlow and the emerging [context engineering](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23contextEngineering) category. Her thesis: AI changes the stakes — the winners will be those who solve for meaning, not simply for calculation.

---

## Architecture Patterns

| Pattern | Optimizes For | Designed For | Key Technology |
|---------|--------------|--------------|----------------|
| [Semantic Layer](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23semanticLayer) | Measurement | Humans (BI) | LookML, MetricFlow, SQL |
| [Ontology](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23ontologyConcept) | Meaning | Systems (AI) | OWL, SKOS, RDF |
| [Knowledge Graph](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23knowledgeGraph) | Relationships | Both | RDF triples, SPARQL |
| [Context Graph](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23contextGraph) | Decision reasoning | AI agents | PKO, Palantir Foundry |
| [Context Engineering](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23contextEngineering) | LLM readiness | AI agents | Prompts, memories, tools |

---

## Ontology Examples

- **[Gene Ontology](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23geneOntology)** — Bioinformatics, 20+ years. Genes, processes, functions, components.
- **[SNOMED CT](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23snomedCT)** — Healthcare, 350K+ concepts, millions of relationships.
- **[PKO](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23proceduralKnowledgeOntology)** — Procedural Knowledge Ontology (Cefriel/Siemens/BOSCH). Procedures vs executions.

---

## FAQ

### [1. Semantic layer vs ontology?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q1)
Semantic layers answer "What is X?" — metrics-first, SQL abstractions. Ontologies support inference — formal concepts, explicit relationships. One optimizes for measurement; the other for meaning.

### [2. Why did semantic layers fail at meaning?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q2)
Knowing revenue = SUM(order_total) WHERE completed doesn't explain why revenue dropped in Q3. Metadata was structural, not semantic.

### [3. Examples of successful ontologies?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q3) · [4. What is a context graph?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q4) · [5. The YAML problem?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q5) · [6. Why AI changes everything?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q6) · [7. Can semantic layers evolve?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q7) · [8. Context engineering?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q8) · [9. Recommendations?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q9) · [10. Metrics-first dead?](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23q10)

---

## HowTo: Evaluating Your Data Architecture for AI (7 Steps)

1. **[Audit your semantic layer's purpose](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23step1)** — "Is this for humans or for AI?"
2. **[Map domain concepts formally](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23step2)** — Ontology engineering, not metric definition
3. **[Capture decision context](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23step3)** — Record why decisions were made
4. **[Invest in knowledge architecture](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23step4)** — A distinct discipline from data engineering
5. **[Evaluate the middle ground](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23step5)** — Context-aware semantic layers
6. **[Learn from existing ontologies](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23step6)** — Gene Ontology, SNOMED CT, Palantir
7. **[Start building now](https://linkeddata.uriburner.com/describe/?url=https%3A%2F%2Fmetadataweekly.substack.com%2Fp%2Fontologies-context-graphs-and-semantic%23step7)** — Don't wait for convergence

---

*Generated 2026-05-26 using kg-generator and rdf-infographic-skill via Claude Opus 4.7 on Claude Code. Entity links via URIBurner. 411 triples.*