Financial planners at Range built agents without coding by choosing repeatable workflows, teaching the agents from real cases, packaging the work as skills, testing edge cases, and assigning owners and reviewers.
2
Agents became useful when they could access connected client data and follow Range's standardised advice processes, rather than working from copied text in a chat window.
3
The bottleneck moved from typing and drafting to keeping skills tested, current, and properly reviewed by financial planning experts.
Summary
Matt Ragone describes how Range's financial planning team adopted agents without an engineering background. The team had already standardised how it handled insurance, investment, retirement, and other recommendations, and client financial data was connected in Range's platform. That made it possible for agents to draft cash flow analyses, insurance reviews, and tax reviews while a planner worked with another client. Ragone's method has five steps: choose a repetitive workflow, teach the agent with real cases and explanations, package the process as a skill, test it on ordinary and edge cases, and ship it with an owner and a reviewer. Every output still requires review by a certified financial planner. After six to eight months, Range had 24 skills across the company, including 17 for financial planning. Ragone says the next operational problem is managing updates, testing, and ownership as the number of skills grows.
Range began with chatbot tasks that still left planners doing the work
The planning team first used AI to redraft emails, check certain tax-code changes, and parse large documents. Those uses saved time, but planners still had to perform the underlying work manually. Ragone describes the old limit as typing speed: even typing very quickly did not remove the need to complete each task. A client call also prevented the planner from working with anyone else during that time. AI made individual steps faster, but it did not yet let the team handle work concurrently.
Agents changed the workflow by drafting work while planners were on calls
With Claude Cowork, Ragone could be on a call with one client while an agent drafted a cash flow analysis, insurance analysis, or tax review for another client. The result was several pieces of work progressing at once. The planner's job then became reviewing and adjusting the draft before it was sent. Ragone is clear that the agent does not remove professional responsibility. The final output has to be checked carefully before it reaches a client.
Useful agents need connected data and standardised advice rules
A good prompt was not enough. The agent had to know where the work lived and how to retrieve the right client data. Range's members connect information about insurance, investments, homes, real estate purchases, and mortgages to the platform. The agent also needed to follow Range's own approach to advice rather than produce generic recommendations. Before agents became available, the roughly 20-person planning team had documented how it handled insurance, investment, and retirement recommendations in Notion pages and Google Docs.
The first workflow should be repetitive, simple, and easy to judge
Ragone advises teams to start with a workflow they perform daily or weekly. He warns against choosing the flashiest or most complex task first. A simple task that takes a lot of time gives the builder immediate feedback about whether the output is good or bad. At Range, examples included cash flow analysis and insurance reviews. The person who knows the work should build the skill, because that person can judge the output while improving it.
Real cases teach the agent the exceptions that process documents miss
Standardisation documents explain how the team normally does the work, but they do not cover every unusual situation. Ragone adds real cases, including cases where he had to correct the agent's work, so the skill can learn from those examples. He also narrates why he is taking particular steps. For example, when uploading a pay stub, he might tell the skill to look for a deduction for group-term life insurance. That extra explanation takes effort at first and reduces repeated correction later.
Skills should ask clarifying questions and enforce review limits
When packaging a workflow into a skill, Ragone has it ask questions that affect the work. A tax review may need a year, while an insurance review may need to focus on homeowners, disability, or whole-life coverage. The skill also needs explicit limits. At Range, no skill can send material directly to a member. A certified financial planner must review it, even though Range has an AI feature called Ry that is CFP and RIA compliant.
Testing requires ordinary cases, edge cases, and repeated revision
Before releasing a skill to the wider team, Ragone runs it on five to ten real cases. Some cases should be standard, while others should expose unusual issues. One or two colleagues can stress-test the skill as well. After each run, the builder updates the skill with what was learned. Ragone also asks the agent how it could improve itself, then applies the suggested changes before shipping. Testing is an ongoing part of the build rather than a final one-time check.
Shipping does not end the work because tax rules and industry practices change. Each skill needs someone responsible for it and someone who reviews it. In a small company, one person may hold both roles. At Range, Ragone might own a skill while a colleague with stronger insurance expertise reviews it. The owner maintains the workflow, while the reviewer adds subject-matter judgment to the output and its updates.
The new bottleneck is managing the growing collection of skills
After six to eight months, Range had 24 skills across the company, with 17 in financial planning. Teams could build on one another's work, and Range was developing Ranger, an agent in Slack that could assist with company and member tasks. Ragone says the constraint has shifted from typing and drafting to deciding who tests and updates each skill and how the company knows it remains current. He expects planners to take on some responsibility as AI agent managers while human advisers continue handling client conversations.