How Financial Planners With Zero Coding Experience Built Their Own Agents - Matt Ragone, Range

Matt Ragone spends much of his day talking to clients about cash flow, taxes, investments, and insurance. He has worked in wealth management for about seven years, with a two-year stint in media during COVID. At Range, an all-in-one wealth management firm, he also leads AI adoption for the planning team. The agents and skills his colleagues use daily were written by the planners themselves. None came to the work with an engineering or coding background.

Ragone represents the planners on Range’s AI Council, a group co-founder David Cusatis conceived so that people using AI across the company could share their work. If member success built a skill that marketing could use, the council gave the two teams a place to discover that. Ragone enjoys the Game of Thrones associations of the name and pictures himself as the warden of the financial planners. His responsibility has a serious constraint. Range operates as a registered investment adviser (RIA), and a recommendation delivered to a client is financial advice. Review has to be part of the workflow.

Range’s planners started by using AI as a chatbot. They could ask it to redraft an email, help check their understanding of a tax code change, or parse a long document. Each use saved a little time. The planner still had to bring the material into the conversation, ask for the work, and carry the answer into the next part of the job. Ragone describes the constraint as his typing speed. Even at a hypothetical 200 or 300 words a minute, he would still have to drive each task himself.

A client call made the cost easier to see. For the 30 or 45 minutes Ragone spent with one client, work for another waited. When Range started using Claude Cowork in late January or early February, he could put a cash flow analysis, an insurance analysis, or a tax review in motion and then take the call. When he hung up, a draft for another member was ready for him to review. Ragone could then check the analysis, make adjustments, and decide whether the advice was ready to send.

Two prerequisites made that handoff possible. First, the agent needed to know where the work lived. Range calls its clients members, and members connect their financial accounts and records to its platform. A planner can work from a picture that includes investments, insurance documents, real estate purchases, and mortgages. The agent needed to retrieve the right document or data from that picture. Having the information in one place solved the problem of repeatedly copying it into a chat window. Knowing which information belonged in a particular analysis was still part of the job.

Second, the agent needed a reference for how Range gives advice. A generic recommendation could be more aggressive than the planning team considered appropriate. Before adopting Cowork, the roughly 20 planners had already standardized how they made insurance, investment, and retirement recommendations. That work originally served a human need. Someone arriving from a large wealth management firm and someone joining from a small RIA could consult the same document to see how Range handled a case.

Those standards lived in Notion pages and Google Docs. The planners could now direct Cowork to consult the relevant document before starting a workflow. The member’s records supplied the facts of the case. The planning document supplied the firm’s method for working through those facts. With both available, the team could concentrate on teaching individual workflows.

Ragone calls the five steps pick, teach, pack, test, and ship. He uses “skill” and “agent” fairly interchangeably in describing the method. The first choice is a job someone does daily or weekly, often enough to know both where the time goes and what good work looks like. Cash flow analyses and insurance reviews fit at Range. Each required substantial manual effort, and an experienced planner could inspect a draft and quickly tell how well it handled the case.

Ragone cautions against starting with the flashiest or most complex assignment. The cost shows up when the first draft comes back. With a familiar, bounded task, the planner can identify a mistake and try a correction straight away. With a complicated assignment, figuring out whether the output is any good can itself take enough time to slow every round of revision.

Ragone’s initial advice was to show the agent one real case. With experience, he would use two to five. The standardization document tells the skill how Range does the work. A finished cash flow analysis shows what a planner actually delivers after applying that method. He adds real cases to the skill’s reference material, including unusual cases and examples in which he had to correct the agent’s work. On a later run, the skill can consult the corrected example when it encounters a similar issue.

As he supplies the documents, Ragone narrates why he is using them. Consider a pay stub uploaded for a cash flow analysis or an insurance review. For the insurance work, he might direct the agent to look for a $3 deduction for group term life insurance. A planner knows why that small entry on a payroll document deserves attention in an insurance review. Uploading the stub alone leaves the agent to infer what the planner wanted it to notice.

The teaching task is to capture that bit of judgment while the practitioner has the case in front of them. It takes longer on the first pass to identify the relevant entry and explain why the document is there. Once the explanation and the example are available to the skill, the next planner running the workflow can benefit from both.

Once a workflow produces good drafts, Ragone packages it as a skill. Range found it worth specifying what should happen when a planner invokes it. The skill can ask whether a tax review covers 2025 or 2026. For a cash flow analysis, it can establish which tax code applies and what state the member lives in. Some of those facts are available from the Range profile. Other questions resolve the purpose of the assignment, which even a detailed profile may leave open.

An insurance review illustrates the difference. Is the member primarily concerned about homeowners insurance? Does the planner want to focus on disability coverage? Is there a whole life policy the member wants reviewed? A couple of questions give the planner a chance to direct the review before the agent spends time preparing the wrong analysis.

Range gives every planning skill the same delivery limit. Its work stops at a draft for the planner. A certified financial planner (CFP) reviews the draft before the member receives anything, and Ragone expects that review to remain part of the process. Range also has a separate built-in AI feature that Ragone describes as CFP- and RIA-compliant. When a member asks for human advice, a planner validates the recommendation.

Before distributing a skill to the rest of the team, Ragone runs it on five to ten real cases. He includes routine examples to see how it handles everyday work and a few edge cases to see what it does when the usual procedure needs adjustment. One or two colleagues may help stress-test it. At this point the question has changed. The examples used in teaching showed the agent what a finished product looked like. The test runs let planners inspect what the packaged skill actually produces.

After each run, Ragone asks Claude to incorporate what it learned into the skill and package it again. Correcting a single draft helps the member whose case is open. Revising the skill gives that correction a chance to affect the next case as well. Before release, he asks Claude what, specifically, it would do to improve the skill further. He uses that answer for another revision, then decides whether the skill is ready for colleagues.

A skill can pass its test cases and still need attention later. Tax codes change, as do practices in the industry. Ragone assigns an owner and a reviewer when a skill ships so the team knows who is responsible for the skill as that happens. In a small firm, one person might fill both roles. Range’s growing planning team can draw on differences in expertise. Ragone might own an insurance skill and ask his colleague Joe to review it because Joe knows the subject better.

That assignment concerns the reusable skill. The CFP review concerns the particular advice a member is about to receive. An insurance specialist can help keep the workflow sound across many runs, while the planner handling a member’s case checks the resulting recommendation against that member’s circumstances.

Roughly six to eight months into the effort, Ragone counted about 24 skills across Range, 17 of them in financial planning. Planners, marketers, and member success staff were building on one another’s work, as the AI Council had intended. Range had also created Ranger, an agent available to people across the company in Slack. Ranger could access the Range dashboard to assist with member tasks. The team was still building out its capabilities.

Within planning, Ragone wants the insurance specialist to iterate on the insurance skill because a specialist will catch details a general practitioner might miss. He compares a CFP to a quarterback who knows when to hand the ball off. The CPA and tax teams contribute the substance of the tax skills, just as a planner can bring in the investment team when a case calls for it. Ragone can help a specialist turn that expertise into questions, references, and revisions without having to supply every judgment himself.

With multiple people running skills, Ragone’s pressing questions are who has tested each one, who is updating it, and how a planner can know whether the version they are using is current. He expects that managing agents may become a substantial part of a planner’s work alongside taking client calls. His ambition is to move from providing advice to three or four people per planner per day to reaching 100. That figure is a goal. What Range had achieved was the ability to prepare one member’s analysis during another member’s call, with a planner still responsible for the advice each receives.