How Financial Planners With Zero Coding Experience Built Their Own Agents - Matt Ragone, Range
Build AI agents around repeatable work: teach them with real examples, test edge cases, and require expert review before use.
- Start with a frequent, simple workflow whose output you can judge quickly.
- Teach the skill with your process, multiple real cases, edge cases, and explanations of your decisions.
- Add clarifying questions and safety limits, test on 5–10 cases, then assign an owner and reviewer.
The speaker walks through a practical build-and-test process with financial-planning examples. · Best for: Nontechnical teams and subject-matter experts building AI workflows
Matt Ragone describes how non-engineers at Range created AI skills for financial-planning work. Their approach relies on documented standards, real examples, iterative testing, and human review—especially important in a regulated business.
Team context and why AI agents were needed0:12
- Range formed an AI Council to share practices across teams.
- Earlier AI use helped draft emails, check tax changes, and parse documents, but staff still did the work manually.
- Claude Co-work enabled agents to prepare analyses while planners served other clients; planners still review and adjust the output.
Give agents access to data and standards4:08
- Agents need to retrieve the right client data from the systems where it lives.
- They also need Range’s documented approach to recommendations, rather than generic advice.
- Existing process standards in Notion and Google Docs provided a reference for skills.
Pick a workflow and teach it6:07
- Use the five-part approach: pick, teach, pack, test, and ship.
- Choose recurring, relatively simple work before attempting a flashy or complex workflow.
- Provide real examples, including edge cases and corrected outputs; explain why you make decisions.
Package skills with questions and limits9:17
- Have the skill ask for missing context, such as the tax year, state, or type of insurance review.
- Specify boundaries: Range’s skills do not send recommendations to members; a CFP reviews the work.
- The speaker also mentions Range’s built-in AI feature, Ry, as CFP- and RIA-compliant.
Test, iterate, and ship responsibly10:16
- Test on 5–10 real cases, including standard and edge cases, and ask colleagues to stress-test the skill.
- Incorporate lessons into the skill and ask it how it could improve before release.
- Assign an owner and reviewer, and keep skills updated as relevant rules and practices change.
Scale through expert ownership12:15
- Range reported 24 skills across the company, including 17 for financial planning, and a company assistant named Ranger in Slack.
- Have subject-matter experts build skills for their specialties, with a separate reviewer where appropriate.
- As AI reduces manual drafting, maintaining, testing, and updating skills becomes a new operational responsibility.
Step by step
- Pick a frequent, manageable workflow with output you can readily evaluate.
- Teach the skill using documented standards and multiple real examples; narrate decisions and include edge cases.
- Package it as a skill; make it ask clarifying questions and state what it must not do.
- Test it on 5–10 real cases, gather colleague feedback, and update it based on results.
- Before release, ask how it could improve; assign an owner and reviewer, then maintain it over time.
Tools & resources mentioned
- Claude Co-work — used to run skills and prepare planning analyses.
- Notion and Google Docs — held the team’s documented recommendation standards.
- Slack — where the company’s Ranger assistant lives.
- Ranger — Range’s developing company-wide assistant for tasks.
- Ry — Range’s built-in AI feature, described by the speaker as CFP- and RIA-compliant.
Notes are AI-generated from the video's transcript and may miss nuance.