JPMorgan’s research team just revealed how they built their multi-agent system “Ask David” and it mirrors a pattern now emerging across AI:
• A supervisor agent orchestrates workflows
• Specialized sub-agents handle retrieval, structured data, and analytics
• An LLM-as-judge layer validates outputs before delivery
• A human-in-the-loop closes the final accuracy gap
This architecture is quickly becoming the default blueprint for production-grade AI systems.