How to Train Your Domain Model
summary
Shows a one-day prototype that retrieves terms and products, generates a recommendation and verifies it with a second LLM (handles a Romanian dialect), demonstrating agents plus guardrails can prototype interactive domain behavior quickly. Uses RAG over S3, agents per bounded context, MCP servers for hot‑swappable tool descriptions and agent traces for auditable decisions. Proposes augmenting aggregates (model maps events to commands) rather than replacing them, and uses model distillation to cut latency and cost. Risks: hallucinations, race conditions, energy and throttling limits.
AI-written summary · about
chapters (14)
- 0:00Initial skepticism of AI
- 2:01Prototyping with AI agents
- 4:43Classifying AI applications
- 7:57Integrating agents with DDD
- 11:27Agent setup and testing
- 14:56Refining agent architecture
- 19:29Ensuring agent correctness
- 22:00The ReAct design pattern
- 24:48Model Context Protocol
- 27:18Performance and distillation
- 30:05Automating agent testing
- 36:51Summary of agent potential
- 42:05Caveats and future outlook
- 44:32Q&A session
related talks
46mArchitecting Efficient Agents with Semantic Routing and Mixture-of-Tools
24mFrom First Experiments to AI Agents: Adoption Experience and Future Challenges
31mDesigning AI Agents with the Model Context Protocol: From Answers to Actions
58mKeynote: Your Agents Are Smart. Your Data Model Isn’t
53mAgents, Tools, and Mcp, Oh My! Next Level AI Concepts for Developers
41mThree agents, three frameworks, one talk
32mScaling Generative AI: Building Production-Ready LLM Applications
37mRAG, Tooling, MCP, Agents ... and What is Missing?