← LOGS

When Financial Planners Write Their Own AI Skills

A financial planner's account of practitioner-built AI skills raises a harder question: who defines, approves, and keeps the professional work correct?

A silhouetted person writing on a glass wall with faint handwritten notes, vertical window frames, and a blurred city skyline in the background.
The skill is new. The judgment behind it took years.

IN BRIEF

In a November 2026 talk, Range financial planner Matt Ragone described how his team turned documented procedures and examples into reusable AI skills. Planners still need to define what a useful draft should contain, review consequential advice before delivery, and keep the instructions current. His account describes an approach, not independently verified quality, compliance, or productivity gains. The useful question is who owns the method, approval, and upkeep when an agent takes on more preparation.

A small deduction on a pay stub can matter to an insurance review. It may be a clue about existing group-term life coverage that a planner should investigate before making a recommendation.

That example comes from Matt Ragone's November 2026 talk for AI Engineer. Ragone, a financial planner at wealth-management firm Range, described how his team had been turning its procedures into reusable AI skills for cash-flow analyses, insurance reviews, and tax-related drafts. Here, a skill means a set of instructions, examples, and reference material that helps an agent prepare a repeatable task.

What caught my attention was the knowledge the planners had to put into those instructions. An agent might flag a payroll deduction, but someone who understands the work has to explain what it could mean, when it matters, and what still needs checking.


The process existed before the skill

Ragone says Range's planners already had standard procedures for insurance, investment, and retirement recommendations, documented in tools such as Notion and Google Docs. They were using AI for individual tasks, including email drafts and document reading, while doing most preparation manually.

His team began giving agents more of that documented process. Ragone says Claude Co-work helped agents prepare drafts while planners attended to other work. He describes the workflow, but doesn't provide measured time savings or evidence that the eventual advice needed less scrutiny.

The existing documentation matters because people can examine it. A procedure can specify which records to collect, which exceptions require follow-up, what questions to ask, and what a usable draft should contain. Without those expectations, an agent could produce a fluent insurance review while overlooking something the team considers essential.

Written instructions can also carry outdated assumptions or mistakes. Their advantage is that someone can challenge and revise them, rather than relying on a vague request to "do a good insurance review."


The people teaching it need to recognise a bad answer

Ragone calls his sequence pick, teach, pack, test, ship. The team starts with repeated tasks whose outputs practitioners understand, supplies examples of completed work, explains why details matter, packages the instructions and references into a skill, and tests it against cases colleagues can review.

Take the pay-stub deduction. Telling an agent to notice it is one thing; a planner has to explain when the clue is relevant and when it might be misleading. Ragone also describes asking for missing context, such as the tax year, state, or focus of an insurance review, before the skill begins. Those inputs can change the analysis.

For testing, Ragone recommends roughly five to ten real cases, including unusual ones, and asking colleagues to challenge the results. His talk provides neither error rates nor an independent comparison with manual work. It describes the testing approach he advocates, not proof that the skills are accurate.

A financial planner may not know how to configure every part of an AI tool. But recognising a convincing-looking wrong answer is exactly the subject expertise needed to define and test what that tool prepares.


Three jobs behind “human review”

Calling an agent's output "human-reviewed" tells us little about the review itself. At least three responsibilities need to be clear.

Defining the method. Someone determines the required inputs, criteria, examples, exceptions, and questions that should stop or redirect preparation. A planner may understand the overall review while a tax specialist is needed for a particular calculation. This expertise shapes the instructions before the agent drafts anything.

Approving the advice. A finished analysis isn't automatically advice fit for a client. Ragone says Range's skills aren't intended to send material directly to a member and that a certified financial planner reviews the output before delivery. That's a reported workflow rule, not independent proof that a technical restriction is enforced or that the process meets regulatory requirements. The described responsibility remains with the professional who approves what goes out.

Maintaining the method. Tax rules change, teams encounter new edge cases, and preferred methods evolve. Ragone says skills need an owner and reviewer, including specialist review where appropriate. Someone must know when the instructions need attention and who can approve changes.

These jobs may belong to different people. The specialist who defines a calculation might not manage access to client records, and the person approving the recommendation might not edit the skill. "Human in the loop" obscures those distinctions unless the responsibilities are named.


Reusable skills still need maintenance

Ragone reports approximately twenty-four skills across Range, seventeen of them in financial planning, from an experiment he describes as six to eight months old. Those are his reported figures, not an independently checked inventory or a measure of performance.

Consider a tax-preparation skill that worked on a case last month. If a relevant rule changes but its instructions don't, the next draft could look complete while using outdated assumptions. That's a hypothetical risk, not an incident Ragone reported at Range.

Once a team relies on reusable skills, someone has to review the instructions, incorporate useful edge cases, control who can change them, and check that revisions are acceptable. Even if the first draft becomes easier to produce, someone still has to review the output and maintain the procedure.

That trade-off could be worthwhile, but the talk doesn't show how much time or money Range has saved or how reliably the skills perform. Assessing that would require evidence about errors, review effort, costs, and the quality of the resulting work.

Ragone also mentions an ambition of serving up to one hundred people a day per planner. It's a target, not observed throughput, and doesn't establish staffing reductions, safer advice, or a productivity gain.


The skill isn't the whole system

Financial planners can define the professional work without being responsible for every technical control around the agent. Securing client data, managing permissions, recording approvals, and testing software failures require other expertise.

A skill can describe what an insurance review should consider. Whether the agent can access a record, send material externally, or bypass an approval depends on its execution environment. I explored that distinction in A Portable Agent Skill Does Not Carry Its Permissions. Instructions alone don't grant or enforce permissions.

Ragone's presentation doesn't establish Range's complete architecture or demonstrate how every restriction is enforced. That doesn't mean those controls are absent. It means the talk can't verify them.

Some preparation tasks may also be handled well by a conventional checklist or template. An agent may offer more when it can assemble information from changing records, ask for missing context, and prepare a draft using the team's method. Whether that advantage outweighs new failure modes needs to be tested.

This approach makes practitioners' judgment easier to inspect and reuse, provided a team also takes responsibility for keeping it accurate.

The agent can prepare the document. It can't inherit the profession's accountability merely by doing so.


// End of transmission. Keep the judgment visible — ZYANE