Keynote: Your Agents Are Smart. Your Data Model Isn’t
summary
Argues that explicit ontologies are the missing semantic contract so agents stop autocompleting policy and actually enforce rules, showing a pharmaceutical demo and a KYC/trust demo that moves from verbose hallucination to auditable denials; uses RDF export from an ontology playground (Fabric), SPARQL queries, a Rust-based symbolic reasoner (MCP server) and Claude as the LLM; McKinsey figure: onboarding a trust takes 2–4 hours, and ontologies produce consistent, checkable outcomes.
AI-written summary · about
chapters (23)
- 0:00Introduction
- 2:14Revisiting ontologies now
- 6:10Industry diagnosis
- 7:34What is an ontology
- 11:10Human and machine models
- 12:53Beyond data models
- 15:38Demo introduction
- 19:59Simple ontologies value
- 21:08Reasoning with and without ontology
- 25:28Interactive ontology building
- 29:37KYC challenges
- 33:48Legal documents as language
- 36:38Discretionary trusts
- 39:45Naive RAG fails for KYC
- 41:30Ontologies for trusts
- 44:44Auditable reasoning
- 46:27Design-time governance
- 48:57Ownership and governance
- 51:05Formalisms and development
- 53:18LLM bias and ontologies
- 54:42Federated ontologies
- 55:44Action plan
- 57:06Summary
related talks
14mWhy Your Agent’s Brain Needs a Playbook: Practical Wins from Using Ontologies
33mKeynote: Boring Wins in AI Agents 🔥 Hard-Earned Lessons
50mKEYNOTE: The agentification of software has a UX problem
30mBuild an AI knowledge fabric for your organisation
17mDon't give your agent the master key
12mEnterprise Agents Have a Structure Problem
01h 01mKeynote Panel: Future of AI agents
50mHow to Train Your Domain Model