We Let Our Financial Planners Write the Agents — Matthew Ragone, Range
This talk outlines a non-coding methodology for Subject Matter Experts (SMEs), specifically financial planners, to build and maintain AI agents, referred to as 'skills.' The process captures institutional knowledge and specific firm workflows, allowing agents to draft complex analyses (e.g., cash flow or insurance reports) while the SME is engaged in other tasks. The core principle is that the agent must replicate *how* the firm operates, not just general best practices. The process emphasizes structured judgment, rigorous testing, and mandatory human review at every stage.
Key takeaways
-
The Agent's Role is Drafting, Not Sending
3:10
A critical safety constraint is that the agent drafts all recommendations, but a human planner must fully review and approve every single recommendation before it is sent to a client. This ensures regulatory compliance and quality control.
-
Skills Capture Institutional Judgment
5:30
A 'skill' is defined as the firm's specific way of performing a piece of work, capturing not just the steps, but the mandatory questions the agent must ask and the limits it must respect. This moves beyond simple prompt engineering.
-
The Five-Step Build Process (Pick, Teach, Pack, Test, Ship)
5:16
The methodology for building a skill requires five steps: 1) Pick a repeatable workflow that is easy to judge; 2) Teach the judgment by narrating real client cases; 3) Pack the knowledge into a skill, defining questions and limits; 4) Test the skill 5-10 times on real cases to find edge cases; and 5) Ship it with a dedicated owner and a separate reviewer.