AI agent skills are written instructions that capture how one team does one specific piece of work, and the people who do that work can build them without writing code. At Range, an all-in-one wealth management company, financial planners built 24 such skills in six months.
How financial planners build AI agent skills without code
Range reported 24 skills across the company six months into its program, 17 of them in planning, according to a 2026 AI Engineer talk by certified financial planner Matthew Ragone. The transcript for that talk lists him as Matt Rigon, the name he goes by; both refer to the same planner, and he also leads AI adoption for the planning team at Range.
Ragone is not an engineer. He spends his days with clients on cash flow, taxes, insurance, retirement and estate questions, and he describes the talk as a method rather than a vendor pitch because the work sits in a regulated business where the output is financial advice and the review step is not optional.
AI agent skills are the firm's own way of doing one piece of work written down so an agent can follow it. The distinction matters because a general model already knows how financial planning works; it does not know which item Range looks at first, which item always gets flagged, or how a specific recommendation is phrased. Once Claude knows how to do a skill, the agent repeats that process on command.
Two conditions had to be true before the method worked at all:
- The agent has to work where the work lives, not in a separate chat window the planner copies and pastes into.
- The agent has to know how Range does the work, not how financial planning works in general.
Why the constraint was never typing speed
The first version of this work looked like every other chatbot trial: paste a client message in, ask for a clearer or friendlier version, and save a few minutes. Ragone describes that stage as useful but manual, since the planner was still doing all of the work and just doing it slightly faster.
The real limit is physical presence. A planner can only be in one place at a time, so a call with one client is time not spent building another client's plan. Work that happens while the planner is somewhere else addresses that limit; faster writing does not.
The current pattern is an agent drafting a full cash flow analysis or an insurance analysis for a different client during a call, followed by full planner review, adjustment, and delivery. The agent sends nothing. A planner reviews every recommendation before a client sees it, and what changed is that the first draft no longer needs someone sitting in front of it.
The five-step build guide for AI agent skills
The build guide runs in five ordered steps, and none of them require writing code, though every one requires knowing the work well. Ragone's argument for that is direct: the people doing the work should be the ones building the instructions, because the instructions are mostly judgment.
| Step | What you do | Why it matters |
|---|---|---|
| 1. Pick | Choose a workflow you repeat and can judge at a glance | If you cannot judge the output quickly, you cannot test the skill |
| 2. Teach | Walk through real past client cases and narrate your reasoning aloud | The judgment, not the steps, is what makes the output the firm's work |
| 3. Pack | Write the questions the agent must ask and the limits it must respect | Steps alone produce something that runs on its own rather than something you can trust |
| 4. Test | Run it 5 to 10 times on real cases before anyone else sees it | Each run surfaces an edge case or a spot where the agent was confidently wrong |
| 5. Ship | Assign an owner who keeps it current and a reviewer other than the author | Someone has to be accountable once 15 people depend on it |
The order carries weight. Picking a workflow whose output can be judged quickly is what makes testing possible at all, and testing is what surfaces the edge cases, the questions the skill should have asked, and the moments where it sounded confident and was wrong.
Pick and teach: start with repeatable, judgeable work
The first instinct is usually the flashiest and most complex work, and Ragone recommends the opposite. Choose something done every week where good output is already recognizable and a bad draft can be spotted quickly, with a bias toward the least complex workflows that require the most manual effort. At Range that meant cash flow analysis and insurance reviews.
If the output cannot be judged quickly, it cannot be tested, and an untested skill cannot be trusted. Two filters decide the candidate: you do it often, and you know how to judge it.
Teaching happens through real cases from plans already delivered, worked start to finish with the reasoning spoken out loud. The narration covers why a document is opened, what is being looked for, and what would cause a pause and a question. The steps are easy to write down; the judgment is what makes the result the firm's work instead of generic advice.
Pack the judgment, then test before shipping
A skill is more than a list of steps. Two elements carry most of the value: the questions the agent must ask before it starts, and the limits it must respect. In Range's world the hard limit is that the agent never sends anything to a client; it drafts, and a planner decides what gets sent or adjusted. The questions cover what the agent needs to know and what it should confirm with the planner instead of guessing.
Writing only the steps produces something that runs on its own. Writing the questions and the limits produces something a professional can trust, because the agent confirms instead of guessing and stops where the firm's rules say it must stop.
Testing means running the skill 5 to 10 times on real cases before anyone else sees it, fixing what each run exposes, saving, and running again. Ragone calls this the step everyone wants to skip. Five to 10 runs sounds like a lot until it is compared with cleaning up after handing a half-finished skill to 15, 20, or 100 colleagues.
Ownership, review, and the quality gap at scale
Shipping a skill means assigning an owner and a reviewer. The owner keeps it current as processes and tax rules change, and Ragone's example of a rule that was right in March being wrong by December sets the cadence. The reviewer checks the skill before the team relies on it, and is someone other than the author, on the same principle that stops a planner grading their own plan.
Once other people depend on a skill, someone has to be accountable for it. That is the line between a clever personal shortcut and something a team can build on.
Adoption spread to sales, product, and marketing without a memo. Ragone reports that all 24 skills were written and are maintained by planners, marketers, and salespeople rather than developers or engineers, and that the skills became more specific once subject-matter experts built for their own niche. Tax knowledge stayed with the tax people, insurance with the insurance people, and review came naturally because the planning team already reviewed each other's plans.
The open challenge is quality at scale. When output requires professional judgment, checking it across a whole company is not a solved problem, and Ragone says there is not much field data on this yet. The bottleneck that has moved is not whether a model can draft a cash flow analysis; it is who writes down how the firm does it, who tests it before it ships, and who owns it once 15 colleagues depend on it.
How to start building AI agent skills in your own team
A team can start with one workflow and finish the loop this week. The sequence below follows the five steps in the order they ran at Range, from choosing the work to naming the person accountable for it afterward.
- Pick a workflow you repeat weekly, where you can spot a bad draft at a glance.
- Walk through several real past cases start to finish and narrate your reasoning out loud, including what would make you stop and ask a question.
- Write the skill, including the questions it must ask before starting and the limits it must never cross.
- Run it 5 to 10 times on real cases, fix what each run exposes, and run it again.
- Ship it with an owner who keeps it current and a reviewer who is not the author.
The constraint to hold onto is the one that separates a drafted recommendation from a sent one. An agent drafts; a person decides. Work that keeps a professional in the decision seat is the work that can survive a review process, a client relationship, and a change in the tax code.
Turning recorded expertise into agent instructions
The step that did the most work in Range's build guide was narration: saying out loud why a document gets opened, what is being looked for, and where you would stop and ask a question. Ragone puts it plainly, calling that narration the whole game. Written knowledge tends to record the procedure and skip the judgment, which is the part a skill actually needs.
If your team already explains its judgment on camera in talks, walkthroughs, or recorded onboarding sessions, that spoken material is closer to a usable skill than most documentation you could write from scratch. The Bootcamp do Dev Doido and Crazystack Typescript cover similar ground for developers working through the same problem of turning tacit practice into explicit instructions.
Skalablog turns a YouTube video into a written article: paste the URL, let the transcript run, and edit the result before it goes out. Material recorded with Gustavo Dev Doido or any other deep technical walkthrough can leave the video frame and become something a team can read, search, and reuse. Skala Blog is where that pipeline starts: paste a YouTube URL, transcribe the video, and generate an article from the transcript.
FAQ
What are AI agent skills?
AI agent skills are written instructions that capture how one organization performs a specific piece of work, including the questions to ask and the limits to respect. At Range, a wealth management firm, a skill covers work such as a cash flow analysis or an insurance review and is followed by an agent on command. The people who do the work write the skill rather than developers.
Do you need to write code to build AI agent skills?
No. Matthew Ragone, a certified financial planner at Range, states that none of the financial planners on the team write code, and that all 24 skills reported at the six-month mark were written and maintained by planners, marketers, and salespeople. The requirement is deep familiarity with the work, not programming.
Which workflow should you automate first?
Start with one you repeat often and can judge quickly, ideally the least complex workflow that consumes the most manual effort. Range began with cash flow analysis and insurance reviews. If you cannot spot a bad draft at a glance, you cannot test the skill, and an untested skill should not be trusted with client-facing work.
How many times should you test a skill before shipping it?
Five to 10 runs on real cases, according to the method described at Range. Each run tends to surface an edge case, a question the skill failed to ask, or a spot where the agent sounded confident and was wrong. That is far less work than correcting a half-finished skill after distributing it to 15 to 100 colleagues.
Does the agent send recommendations to clients?
No. In Range's workflow the agent drafts and a planner reviews and adjusts every recommendation before a client sees it. That limit is written into the skill itself, alongside the questions the agent must confirm rather than guess, and it is what keeps professional review in the loop.
Who should own a skill after it ships?
An owner keeps it current, and a reviewer other than the author checks it before the team depends on it. Ragone notes that firm processes and tax rules change, so a skill that was right in March can be wrong by December. Ownership is what separates a personal shortcut from shared infrastructure.
Why did skills spread beyond the planning team?
The spread happened without a mandate or a memo. Ragone reports adoption in sales, product, and marketing, and says the skills became more specific once subject-matter experts built for their own niche. The habit of reviewing each other's work carried over from plan review.
What changes once non-engineers build for their own niche?
Two things, in Range's account: the skills get more specific, and review happens naturally. The person who knows insurance best builds the insurance skill, and the tax skill goes to the person who knows tax law. Reviewing each other's skills becomes an extension of reviewing each other's plans.
What is still hard about this approach?
Checking quality at scale remains the open problem. When output requires professional judgment, a company cannot simply count errors, and Ragone says there is little field data on this yet as of the 2026 talk. The method covers building, testing, and ownership, but the measure of whether a skill is working across a whole team is not settled.
FAQ
- What are AI agent skills? AI agent skills are written instructions that capture how one organization performs a specific piece of work, including the questions to ask and the limits to respect. At Range, a wealth management firm, a skill covers work such as a cash flow analysis or an insurance review and is followed by an agent on command. The people who do the work write the skill rather than developers.
- Do you need to write code to build AI agent skills? No. Matthew Ragone, a certified financial planner at Range, states that none of the financial planners on the team write code, and that all 24 skills reported at the six-month mark were written and maintained by planners, marketers, and salespeople. The requirement is deep familiarity with the work, not programming.
- Which workflow should you automate first? Start with one you repeat often and can judge quickly, ideally the least complex workflow that consumes the most manual effort. Range began with cash flow analysis and insurance reviews. If you cannot spot a bad draft at a glance, you cannot test the skill, and an untested skill should not be trusted with client-facing work.
- How many times should you test a skill before shipping it? Five to 10 runs on real cases, according to the method described at Range. Each run tends to surface an edge case, a question the skill failed to ask, or a spot where the agent sounded confident and was wrong. That is far less work than correcting a half-finished skill after distributing it to 15 to 100 colleagues.
- Does the agent send recommendations to clients? No. In Range's workflow the agent drafts and a planner reviews and adjusts every recommendation before a client sees it. That limit is written into the skill itself, alongside the questions the agent must confirm rather than guess, and it is what keeps professional review in the loop.
- Who should own a skill after it ships? An owner keeps it current, and a reviewer other than the author checks it before the team depends on it. Ragone notes that firm processes and tax rules change, so a skill that was right in March can be wrong by December. Ownership is what separates a personal shortcut from shared infrastructure.
- Why did skills spread beyond the planning team? People outside planning saw what the planners were doing and built their own, with no memo or mandate involved. Ragone reports adoption in sales, product, and marketing, and says the skills became more specific once subject-matter experts built for their own niche. The habit of reviewing each other's work carried over from plan review.
- What changes once non-engineers build for their own niche? Two things, in Range's account: the skills get more specific, and review happens naturally. The person who knows insurance best builds the insurance skill, and the tax skill goes to the person who knows tax law. Reviewing each other's skills becomes an extension of reviewing each other's plans.
- What is still hard about this approach? Checking quality at scale remains the open problem. When output requires professional judgment, a company cannot simply count errors, and Ragone says there is little field data on this yet. The method covers building, testing, and ownership, but the measure of whether a skill is working across a whole team is not settled.
A written form for spoken expertise
The whole method rests on one observation: the value in a skill is judgment, and judgment is easiest to capture out loud while you are doing the work. Ragone recorded his explanation on camera, and that recording carried the detail a written procedure would have left out.
If your team holds knowledge in the same place, in recorded calls, internal walkthroughs, or conference talks, Skalablog takes a YouTube URL, transcribes the video, and turns it into an article draft you can edit and publish under your own name. That is the whole flow: YouTube video, transcript, article.
Fork this article
Start a new branch from the same video, shaped your way. You keep the credit; the original keeps the attribution.
A fork in another language is filed as a translation of this article, so the two pages point at each other. You can unlink it later from the editor.
0/240
You are creating
- Format
- For
- Language
- Source
- Your angle
You will be asked to sign in before it is generated.
Buy credits